FLOPs全称是floating point operations的缩写,翻译过来是浮点运算数,理解为计算量,常用来衡量算法或深度学习模型的计算复杂度。
关于计算FLOPs值的函数,网上相关的博客很多,但是能用的很少,下面这个函数是我实际使用过可行的函数,用来计算keras模型的FLOPs值。
方法一:
# 浮点运行次数
# FLOPS:注意全大写,是floating point operations per second的缩写,意指每秒浮点运算次数,理解为计算速度。是一个衡量硬件性能的指标。
# FLOPs:注意s小写,是floating point operations的缩写(s表复数),意指浮点运算数,理解为计算量。可以用来衡量算法/模型的复杂度。
# In TF 2.x you have to use tf.compat.v1.RunMetadata instead of tf.RunMetadata
# To work your code in TF 2.1.0, i have made all necessary changes that are compliant to TF 2.x
# print(tf.__version__)
import tensorflow as tf
# 必须要下面这行代码
tf.compat.v1.disable_eager_execution()
print(tf.__version__)
# 我自己使用的函数
def get_flops_params():
sess = tf.compat.v1.Session()
graph = sess.graph
flops = tf.compat.v1.profiler.profile(graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.float_operation())
params = tf.compat.v1.profiler.profile(graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.trainable_variables_parameter())
print('FLOPs: {}; Trainable params: {}'.format(flops.total_float_ops, params.total_parameters))
# 网上推荐的
# sess = tf.compat.v1.Session()
# graph = sess.graph
# stats_graph(graph)
def stats_graph(graph):
flops = tf.compat.v1.profiler.profile(graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.float_operation())
# print('FLOPs: {}'.format(flops.total_float_ops))
params = tf.compat.v1.profiler.profile(graph, options=tf.compat.v1.profiler.ProfileOptionBuilder.trainable_variables_parameter())
# print('Trainable params: {}'.format(params.total_parameters))
print('FLOPs: {}; Trainable params: {}'.format(flops.total_float_ops, params.total_parameters))
def get_flops(model):
run_meta = tf.compat.v1.RunMetadata()
opts = tf.compat.v1.profiler.ProfileOptionBuilder.float_operation()
# We use the Keras session graph in the call to the profiler.
flops = tf.compat.v1.profiler.profile(graph=tf.compat.v1.keras.backend.get_session().graph, run_meta=run_meta, cmd='op', options=opts)
return flops.total_float_ops # Prints the "flops" of the model.
# 必须使用tensorflow中的keras才能够获取到FLOPs, 模型中的各个函数都必须使用tensorflow.keras中的函数,和keras混用会报错
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense
from tensorflow.keras.models import Sequential
model = Sequential()
model.add(Conv2D(filters=64, kernel_size=(3, 3), input_shape=(28, 28, 1), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(units=100, activation='relu'))
model.add(Dense(units=10, activation='softmax'))
# 获取模型每一层的参数详情
model.summary()
# 获取模型浮点运算总次数和模型的总参数
get_flops_params()
此代码情况下,实测MobileNetv2计算量误差较大:
2021-04-18 21:17:58.053772: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-04-18 21:17:58.054691: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1261] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-04-18 21:17:58.054932: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267]
2021-04-18 21:17:58.055070: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
WARNING:tensorflow:From D:\Anaconda3\envs\wen\lib\site-packages\tensorflow\python\profiler\internal\flops_registry.py:142: tensor_shape_from_node_def_name (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.tensor_shape_from_node_def_name`
[0418 21:17:58] From D:\Anaconda3\envs\wen\lib\site-packages\tensorflow\python\profiler\internal\flops_registry.py:142: tensor_shape_from_node_def_name (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.tensor_shape_from_node_def_name`
62 ops no flops stats due to incomplete shapes.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
=========================Options=============================
-max_depth 10000
-min_bytes 0
-min_peak_bytes 0
-min_residual_bytes 0
-min_output_bytes 0
-min_micros 0
-min_accelerator_micros 0
-min_cpu_micros 0
-min_params 0
-min_float_ops 1
-min_occurrence 0
-step -1
-order_by float_ops
-account_type_regexes .*
-start_name_regexes .*
-trim_name_regexes
-show_name_regexes .*
-hide_name_regexes
-account_displayed_op_only true
-select float_ops
-output stdout:
==================Model Analysis Report======================
Doc:
scope: The nodes in the model graph are organized by their names, which is hierarchical like filesystem.
flops: Number of float operations. Note: Please read the implementation for the math behind it.
Profile:
node name | # float_ops
_TFProfRoot (--/7.01m flops)
predictions/kernel/Initializer/random_uniform (1.28m/2.56m flops)
predictions/kernel/Initializer/random_uniform/mul (1.28m/1.28m flops)
predictions/kernel/Initializer/random_uniform/sub (1/1 flops)
Conv_1/kernel/Initializer/random_uniform (409.60k/819.20k flops)
Conv_1/kernel/Initializer/random_uniform/mul (409.60k/409.60k flops)
Conv_1/kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_project/kernel/Initializer/random_uniform (307.20k/614.40k flops)
block_16_project/kernel/Initializer/random_uniform/mul (307.20k/307.20k flops)
block_16_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_project/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_14_project/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_14_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_14_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_14_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_project/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_15_project/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_15_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_15_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_15_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_16_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_16_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_project/kernel/Initializer/random_uniform (92.16k/184.32k flops)
block_13_project/kernel/Initializer/random_uniform/mul (92.16k/92.16k flops)
block_13_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_project/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_12_project/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_12_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_13_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_13_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_12_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_12_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_11_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_11_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_project/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_11_project/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_11_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_project/kernel/Initializer/random_uniform (36.86k/73.73k flops)
block_10_project/kernel/Initializer/random_uniform/mul (36.86k/36.86k flops)
block_10_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_8_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_8_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_8_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_8_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_9_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_9_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_7_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_7_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_9_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_9_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_7_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_7_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_10_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_10_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_project/kernel/Initializer/random_uniform (12.29k/24.58k flops)
block_6_project/kernel/Initializer/random_uniform/mul (12.29k/12.29k flops)
block_6_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_6_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_6_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_project/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_4_project/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_4_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_5_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_5_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_project/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_5_project/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_5_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_4_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_4_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_project/kernel/Initializer/random_uniform (4.61k/9.22k flops)
block_3_project/kernel/Initializer/random_uniform/mul (4.61k/4.61k flops)
block_3_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_expand/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_2_expand/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_2_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_project/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_2_project/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_2_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_expand/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_3_expand/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_3_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_project/kernel/Initializer/random_uniform (2.30k/4.61k flops)
block_1_project/kernel/Initializer/random_uniform/mul (2.30k/2.30k flops)
block_1_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_expand/kernel/Initializer/random_uniform (1.54k/3.07k flops)
block_1_expand/kernel/Initializer/random_uniform/mul (1.54k/1.54k flops)
block_1_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform (1.30k/2.59k flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.30k/1.30k flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform (1.30k/2.59k flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.30k/1.30k flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
Conv1/kernel/Initializer/random_uniform (864/1.73k flops)
Conv1/kernel/Initializer/random_uniform/mul (864/864 flops)
Conv1/kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform (864/1.73k flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform/mul (864/864 flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg/mul (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg/sub_1 (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg_1/mul (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg_1/sub_1 (1.28k/1.28k flops)
expanded_conv_project/kernel/Initializer/random_uniform (512/1.02k flops)
expanded_conv_project/kernel/Initializer/random_uniform/mul (512/512 flops)
expanded_conv_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_15_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_15_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_14_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_14_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_14_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_15_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_14_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_15_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_16_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_16_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_16_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform (288/577 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform/mul (288/288 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_13_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_13_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_11_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_11_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_11_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_11_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_13_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_12_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_12_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_12_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_12_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_8_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_8_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_8_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_10_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_10_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_10_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_10_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_9_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_9_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_7_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_7_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_16_project_BN/AssignMovingAvg_1/sub_1 (320/320 flops)
block_16_project_BN/AssignMovingAvg_1/mul (320/320 flops)
block_16_project_BN/AssignMovingAvg/sub_1 (320/320 flops)
block_16_project_BN/AssignMovingAvg/mul (320/320 flops)
block_4_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_5_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_6_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_5_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_6_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_6_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_5_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_5_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_6_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_4_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_4_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_4_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_4_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_14_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_14_project_BN/AssignMovingAvg/mul (160/160 flops)
block_13_project_BN/AssignMovingAvg/mul (160/160 flops)
block_13_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_13_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_15_project_BN/AssignMovingAvg/mul (160/160 flops)
block_13_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_15_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_14_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_15_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_14_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_15_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_3_expand_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_3_expand_BN/AssignMovingAvg_1/mul (144/144 flops)
block_3_expand_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg/mul (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg_1/mul (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_2_expand_BN/AssignMovingAvg/mul (144/144 flops)
block_2_expand_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_2_expand_BN/AssignMovingAvg_1/mul (144/144 flops)
block_2_expand_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg/mul (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg_1/mul (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_3_expand_BN/AssignMovingAvg/mul (144/144 flops)
block_1_depthwise_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_1_expand_BN/AssignMovingAvg_1/mul (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg/mul (96/96 flops)
block_11_project_BN/AssignMovingAvg/mul (96/96 flops)
block_1_expand_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_11_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_1_expand_BN/AssignMovingAvg/mul (96/96 flops)
block_11_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_11_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_1_expand_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg/mul (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg_1/mul (96/96 flops)
block_10_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_10_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_10_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_10_project_BN/AssignMovingAvg/mul (96/96 flops)
block_6_project_BN/AssignMovingAvg/mul (64/64 flops)
block_6_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_7_project_BN/AssignMovingAvg/mul (64/64 flops)
block_7_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_8_project_BN/AssignMovingAvg/mul (64/64 flops)
block_8_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_8_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_8_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_7_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_7_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_6_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_6_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_9_project_BN/AssignMovingAvg/mul (64/64 flops)
block_9_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_9_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_9_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_5_project_BN/AssignMovingAvg/mul (32/32 flops)
block_4_project_BN/AssignMovingAvg/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
block_4_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
bn_Conv1/AssignMovingAvg/sub_1 (32/32 flops)
block_4_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_4_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_3_project_BN/AssignMovingAvg/mul (32/32 flops)
block_3_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
block_3_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_3_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
bn_Conv1/AssignMovingAvg/mul (32/32 flops)
bn_Conv1/AssignMovingAvg_1/mul (32/32 flops)
bn_Conv1/AssignMovingAvg_1/sub_1 (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/mul (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/sub_1 (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/mul (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_1_project_BN/AssignMovingAvg/mul (24/24 flops)
block_2_project_BN/AssignMovingAvg/mul (24/24 flops)
block_1_project_BN/AssignMovingAvg/sub_1 (24/24 flops)
block_2_project_BN/AssignMovingAvg_1/sub_1 (24/24 flops)
block_1_project_BN/AssignMovingAvg_1/mul (24/24 flops)
block_2_project_BN/AssignMovingAvg_1/mul (24/24 flops)
block_2_project_BN/AssignMovingAvg/sub_1 (24/24 flops)
block_1_project_BN/AssignMovingAvg_1/sub_1 (24/24 flops)
expanded_conv_project_BN/AssignMovingAvg/mul (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg/sub_1 (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg_1/mul (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg_1/sub_1 (16/16 flops)
block_16_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_5_project_BN/AssignMovingAvg/sub (1/1 flops)
block_5_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_project_BN/AssignMovingAvg/sub (1/1 flops)
block_6_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_15_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_6_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_14_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_14_project_BN/AssignMovingAvg/sub (1/1 flops)
block_14_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_14_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_14_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_14_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_6_project_BN/AssignMovingAvg_1/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg/sub (1/1 flops)
block_13_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_13_project_BN/AssignMovingAvg/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_13_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_13_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_7_expand_BN/AssignMovingAvg/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg_1/sub (1/1 flops)
block_12_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_10_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_2_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_2_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_project_BN/AssignMovingAvg/sub (1/1 flops)
block_3_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_2_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_7_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_2_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_project_BN/AssignMovingAvg/sub (1/1 flops)
block_3_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_8_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_12_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_8_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_2_project_BN/AssignMovingAvg/sub (1/1 flops)
block_8_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_11_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_8_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_project_BN/AssignMovingAvg/sub (1/1 flops)
block_3_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_2_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_8_project_BN/AssignMovingAvg/sub (1/1 flops)
block_1_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_project_BN/AssignMovingAvg/sub (1/1 flops)
block_8_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_9_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_11_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_9_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_project_BN/AssignMovingAvg/sub (1/1 flops)
block_10_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_9_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_1_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_10_project_BN/AssignMovingAvg/sub (1/1 flops)
block_9_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_10_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_9_project_BN/AssignMovingAvg/sub (1/1 flops)
block_4_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_project_BN/AssignMovingAvg/sub (1/1 flops)
block_9_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
bn_Conv1/AssignMovingAvg/sub (1/1 flops)
block_11_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_5_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
bn_Conv1/AssignMovingAvg_1/sub (1/1 flops)
block_1_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_10_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_5_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_16_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_16_project_BN/AssignMovingAvg/sub (1/1 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_5_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_10_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_16_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_project_BN/AssignMovingAvg/sub (1/1 flops)
block_16_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_5_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_16_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
loss/mul (1/1 flops)
loss/predictions_loss/softmax_cross_entropy_with_logits/Sub (1/1 flops)
loss/predictions_loss/softmax_cross_entropy_with_logits/Sub_1 (1/1 flops)
loss/predictions_loss/softmax_cross_entropy_with_logits/Sub_2 (1/1 flops)
block_6_project_BN/AssignMovingAvg/sub (1/1 flops)
======================End of Report==========================
62 ops no flops stats due to incomplete shapes.
Incomplete shape.
Incomplete shape.
=========================Options=============================
-max_depth 10000
-min_bytes 0
-min_peak_bytes 0
-min_residual_bytes 0
-min_output_bytes 0
-min_micros 0
-min_accelerator_micros 0
-min_cpu_micros 0
-min_params 0
-min_float_ops 0
-min_occurrence 0
-step -1
-order_by name
-account_type_regexes _trainable_variables
-start_name_regexes .*
-trim_name_regexes
-show_name_regexes .*
-hide_name_regexes
-account_displayed_op_only true
-select params
-output stdout:
==================Model Analysis Report======================
Doc:
scope: The nodes in the model graph are organized by their names, which is hierarchical like filesystem.
param: Number of parameters (in the Variable).
Profile:
node name | # parameters
_TFProfRoot (--/3.50m params)
Conv1 (--/864 params)
Conv1/kernel (3x3x3x32, 864/864 params)
Conv_1 (--/409.60k params)
Conv_1/kernel (1x1x320x1280, 409.60k/409.60k params)
Conv_1_bn (--/2.56k params)
Conv_1_bn/beta (1280, 1.28k/1.28k params)
Conv_1_bn/gamma (1280, 1.28k/1.28k params)
block_10_depthwise (--/3.46k params)
block_10_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_10_depthwise_BN (--/768 params)
block_10_depthwise_BN/beta (384, 384/384 params)
block_10_depthwise_BN/gamma (384, 384/384 params)
block_10_expand (--/24.58k params)
block_10_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_10_expand_BN (--/768 params)
block_10_expand_BN/beta (384, 384/384 params)
block_10_expand_BN/gamma (384, 384/384 params)
block_10_project (--/36.86k params)
block_10_project/kernel (1x1x384x96, 36.86k/36.86k params)
block_10_project_BN (--/192 params)
block_10_project_BN/beta (96, 96/96 params)
block_10_project_BN/gamma (96, 96/96 params)
block_11_depthwise (--/5.18k params)
block_11_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_11_depthwise_BN (--/1.15k params)
block_11_depthwise_BN/beta (576, 576/576 params)
block_11_depthwise_BN/gamma (576, 576/576 params)
block_11_expand (--/55.30k params)
block_11_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_11_expand_BN (--/1.15k params)
block_11_expand_BN/beta (576, 576/576 params)
block_11_expand_BN/gamma (576, 576/576 params)
block_11_project (--/55.30k params)
block_11_project/kernel (1x1x576x96, 55.30k/55.30k params)
block_11_project_BN (--/192 params)
block_11_project_BN/beta (96, 96/96 params)
block_11_project_BN/gamma (96, 96/96 params)
block_12_depthwise (--/5.18k params)
block_12_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_12_depthwise_BN (--/1.15k params)
block_12_depthwise_BN/beta (576, 576/576 params)
block_12_depthwise_BN/gamma (576, 576/576 params)
block_12_expand (--/55.30k params)
block_12_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_12_expand_BN (--/1.15k params)
block_12_expand_BN/beta (576, 576/576 params)
block_12_expand_BN/gamma (576, 576/576 params)
block_12_project (--/55.30k params)
block_12_project/kernel (1x1x576x96, 55.30k/55.30k params)
block_12_project_BN (--/192 params)
block_12_project_BN/beta (96, 96/96 params)
block_12_project_BN/gamma (96, 96/96 params)
block_13_depthwise (--/5.18k params)
block_13_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_13_depthwise_BN (--/1.15k params)
block_13_depthwise_BN/beta (576, 576/576 params)
block_13_depthwise_BN/gamma (576, 576/576 params)
block_13_expand (--/55.30k params)
block_13_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_13_expand_BN (--/1.15k params)
block_13_expand_BN/beta (576, 576/576 params)
block_13_expand_BN/gamma (576, 576/576 params)
block_13_project (--/92.16k params)
block_13_project/kernel (1x1x576x160, 92.16k/92.16k params)
block_13_project_BN (--/320 params)
block_13_project_BN/beta (160, 160/160 params)
block_13_project_BN/gamma (160, 160/160 params)
block_14_depthwise (--/8.64k params)
block_14_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_14_depthwise_BN (--/1.92k params)
block_14_depthwise_BN/beta (960, 960/960 params)
block_14_depthwise_BN/gamma (960, 960/960 params)
block_14_expand (--/153.60k params)
block_14_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_14_expand_BN (--/1.92k params)
block_14_expand_BN/beta (960, 960/960 params)
block_14_expand_BN/gamma (960, 960/960 params)
block_14_project (--/153.60k params)
block_14_project/kernel (1x1x960x160, 153.60k/153.60k params)
block_14_project_BN (--/320 params)
block_14_project_BN/beta (160, 160/160 params)
block_14_project_BN/gamma (160, 160/160 params)
block_15_depthwise (--/8.64k params)
block_15_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_15_depthwise_BN (--/1.92k params)
block_15_depthwise_BN/beta (960, 960/960 params)
block_15_depthwise_BN/gamma (960, 960/960 params)
block_15_expand (--/153.60k params)
block_15_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_15_expand_BN (--/1.92k params)
block_15_expand_BN/beta (960, 960/960 params)
block_15_expand_BN/gamma (960, 960/960 params)
block_15_project (--/153.60k params)
block_15_project/kernel (1x1x960x160, 153.60k/153.60k params)
block_15_project_BN (--/320 params)
block_15_project_BN/beta (160, 160/160 params)
block_15_project_BN/gamma (160, 160/160 params)
block_16_depthwise (--/8.64k params)
block_16_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_16_depthwise_BN (--/1.92k params)
block_16_depthwise_BN/beta (960, 960/960 params)
block_16_depthwise_BN/gamma (960, 960/960 params)
block_16_expand (--/153.60k params)
block_16_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_16_expand_BN (--/1.92k params)
block_16_expand_BN/beta (960, 960/960 params)
block_16_expand_BN/gamma (960, 960/960 params)
block_16_project (--/307.20k params)
block_16_project/kernel (1x1x960x320, 307.20k/307.20k params)
block_16_project_BN (--/640 params)
block_16_project_BN/beta (320, 320/320 params)
block_16_project_BN/gamma (320, 320/320 params)
block_1_depthwise (--/864 params)
block_1_depthwise/depthwise_kernel (3x3x96x1, 864/864 params)
block_1_depthwise_BN (--/192 params)
block_1_depthwise_BN/beta (96, 96/96 params)
block_1_depthwise_BN/gamma (96, 96/96 params)
block_1_expand (--/1.54k params)
block_1_expand/kernel (1x1x16x96, 1.54k/1.54k params)
block_1_expand_BN (--/192 params)
block_1_expand_BN/beta (96, 96/96 params)
block_1_expand_BN/gamma (96, 96/96 params)
block_1_project (--/2.30k params)
block_1_project/kernel (1x1x96x24, 2.30k/2.30k params)
block_1_project_BN (--/48 params)
block_1_project_BN/beta (24, 24/24 params)
block_1_project_BN/gamma (24, 24/24 params)
block_2_depthwise (--/1.30k params)
block_2_depthwise/depthwise_kernel (3x3x144x1, 1.30k/1.30k params)
block_2_depthwise_BN (--/288 params)
block_2_depthwise_BN/beta (144, 144/144 params)
block_2_depthwise_BN/gamma (144, 144/144 params)
block_2_expand (--/3.46k params)
block_2_expand/kernel (1x1x24x144, 3.46k/3.46k params)
block_2_expand_BN (--/288 params)
block_2_expand_BN/beta (144, 144/144 params)
block_2_expand_BN/gamma (144, 144/144 params)
block_2_project (--/3.46k params)
block_2_project/kernel (1x1x144x24, 3.46k/3.46k params)
block_2_project_BN (--/48 params)
block_2_project_BN/beta (24, 24/24 params)
block_2_project_BN/gamma (24, 24/24 params)
block_3_depthwise (--/1.30k params)
block_3_depthwise/depthwise_kernel (3x3x144x1, 1.30k/1.30k params)
block_3_depthwise_BN (--/288 params)
block_3_depthwise_BN/beta (144, 144/144 params)
block_3_depthwise_BN/gamma (144, 144/144 params)
block_3_expand (--/3.46k params)
block_3_expand/kernel (1x1x24x144, 3.46k/3.46k params)
block_3_expand_BN (--/288 params)
block_3_expand_BN/beta (144, 144/144 params)
block_3_expand_BN/gamma (144, 144/144 params)
block_3_project (--/4.61k params)
block_3_project/kernel (1x1x144x32, 4.61k/4.61k params)
block_3_project_BN (--/64 params)
block_3_project_BN/beta (32, 32/32 params)
block_3_project_BN/gamma (32, 32/32 params)
block_4_depthwise (--/1.73k params)
block_4_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_4_depthwise_BN (--/384 params)
block_4_depthwise_BN/beta (192, 192/192 params)
block_4_depthwise_BN/gamma (192, 192/192 params)
block_4_expand (--/6.14k params)
block_4_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_4_expand_BN (--/384 params)
block_4_expand_BN/beta (192, 192/192 params)
block_4_expand_BN/gamma (192, 192/192 params)
block_4_project (--/6.14k params)
block_4_project/kernel (1x1x192x32, 6.14k/6.14k params)
block_4_project_BN (--/64 params)
block_4_project_BN/beta (32, 32/32 params)
block_4_project_BN/gamma (32, 32/32 params)
block_5_depthwise (--/1.73k params)
block_5_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_5_depthwise_BN (--/384 params)
block_5_depthwise_BN/beta (192, 192/192 params)
block_5_depthwise_BN/gamma (192, 192/192 params)
block_5_expand (--/6.14k params)
block_5_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_5_expand_BN (--/384 params)
block_5_expand_BN/beta (192, 192/192 params)
block_5_expand_BN/gamma (192, 192/192 params)
block_5_project (--/6.14k params)
block_5_project/kernel (1x1x192x32, 6.14k/6.14k params)
block_5_project_BN (--/64 params)
block_5_project_BN/beta (32, 32/32 params)
block_5_project_BN/gamma (32, 32/32 params)
block_6_depthwise (--/1.73k params)
block_6_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_6_depthwise_BN (--/384 params)
block_6_depthwise_BN/beta (192, 192/192 params)
block_6_depthwise_BN/gamma (192, 192/192 params)
block_6_expand (--/6.14k params)
block_6_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_6_expand_BN (--/384 params)
block_6_expand_BN/beta (192, 192/192 params)
block_6_expand_BN/gamma (192, 192/192 params)
block_6_project (--/12.29k params)
block_6_project/kernel (1x1x192x64, 12.29k/12.29k params)
block_6_project_BN (--/128 params)
block_6_project_BN/beta (64, 64/64 params)
block_6_project_BN/gamma (64, 64/64 params)
block_7_depthwise (--/3.46k params)
block_7_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_7_depthwise_BN (--/768 params)
block_7_depthwise_BN/beta (384, 384/384 params)
block_7_depthwise_BN/gamma (384, 384/384 params)
block_7_expand (--/24.58k params)
block_7_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_7_expand_BN (--/768 params)
block_7_expand_BN/beta (384, 384/384 params)
block_7_expand_BN/gamma (384, 384/384 params)
block_7_project (--/24.58k params)
block_7_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_7_project_BN (--/128 params)
block_7_project_BN/beta (64, 64/64 params)
block_7_project_BN/gamma (64, 64/64 params)
block_8_depthwise (--/3.46k params)
block_8_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_8_depthwise_BN (--/768 params)
block_8_depthwise_BN/beta (384, 384/384 params)
block_8_depthwise_BN/gamma (384, 384/384 params)
block_8_expand (--/24.58k params)
block_8_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_8_expand_BN (--/768 params)
block_8_expand_BN/beta (384, 384/384 params)
block_8_expand_BN/gamma (384, 384/384 params)
block_8_project (--/24.58k params)
block_8_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_8_project_BN (--/128 params)
block_8_project_BN/beta (64, 64/64 params)
block_8_project_BN/gamma (64, 64/64 params)
block_9_depthwise (--/3.46k params)
block_9_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_9_depthwise_BN (--/768 params)
block_9_depthwise_BN/beta (384, 384/384 params)
block_9_depthwise_BN/gamma (384, 384/384 params)
block_9_expand (--/24.58k params)
block_9_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_9_expand_BN (--/768 params)
block_9_expand_BN/beta (384, 384/384 params)
block_9_expand_BN/gamma (384, 384/384 params)
block_9_project (--/24.58k params)
block_9_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_9_project_BN (--/128 params)
block_9_project_BN/beta (64, 64/64 params)
block_9_project_BN/gamma (64, 64/64 params)
bn_Conv1 (--/64 params)
bn_Conv1/beta (32, 32/32 params)
bn_Conv1/gamma (32, 32/32 params)
expanded_conv_depthwise (--/288 params)
expanded_conv_depthwise/depthwise_kernel (3x3x32x1, 288/288 params)
expanded_conv_depthwise_BN (--/64 params)
expanded_conv_depthwise_BN/beta (32, 32/32 params)
expanded_conv_depthwise_BN/gamma (32, 32/32 params)
expanded_conv_project (--/512 params)
expanded_conv_project/kernel (1x1x32x16, 512/512 params)
expanded_conv_project_BN (--/32 params)
expanded_conv_project_BN/beta (16, 16/16 params)
expanded_conv_project_BN/gamma (16, 16/16 params)
predictions (--/1.28m params)
predictions/bias (1000, 1.00k/1.00k params)
predictions/kernel (1280x1000, 1.28m/1.28m params)
======================End of Report==========================
Incomplete shape.
Incomplete shape.
FLOPs: 7,007,905; Trainable params: 3,504,872
以上是keras的application中给出的代码实现,以下是自实现MobileNet的结果测试:
Use `tf.compat.v1.graph_util.tensor_shape_from_node_def_name`
[0418 21:45:18] From D:\Anaconda3\envs\wen\lib\site-packages\tensorflow\python\profiler\internal\flops_registry.py:142: tensor_shape_from_node_def_name (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.tensor_shape_from_node_def_name`
62 ops no flops stats due to incomplete shapes.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
=========================Options=============================
-max_depth 10000
-min_bytes 0
-min_peak_bytes 0
-min_residual_bytes 0
-min_output_bytes 0
-min_micros 0
-min_accelerator_micros 0
-min_cpu_micros 0
-min_params 0
-min_float_ops 1
-min_occurrence 0
-step -1
-order_by float_ops
-account_type_regexes .*
-start_name_regexes .*
-trim_name_regexes
-show_name_regexes .*
-hide_name_regexes
-account_displayed_op_only true
-select float_ops
-output stdout:
==================Model Analysis Report======================
Doc:
scope: The nodes in the model graph are organized by their names, which is hierarchical like filesystem.
flops: Number of float operations. Note: Please read the implementation for the math behind it.
Profile:
node name | # float_ops
_TFProfRoot (--/4.57m flops)
Conv_1/kernel/Initializer/random_uniform (409.60k/819.20k flops)
Conv_1/kernel/Initializer/random_uniform/mul (409.60k/409.60k flops)
Conv_1/kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_project/kernel/Initializer/random_uniform (307.20k/614.40k flops)
block_16_project/kernel/Initializer/random_uniform/mul (307.20k/307.20k flops)
block_16_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_project/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_15_project/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_15_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_16_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_16_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_project/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_14_project/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_14_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_15_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_15_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_expand/kernel/Initializer/random_uniform (153.60k/307.20k flops)
block_14_expand/kernel/Initializer/random_uniform/mul (153.60k/153.60k flops)
block_14_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_project/kernel/Initializer/random_uniform (92.16k/184.32k flops)
block_13_project/kernel/Initializer/random_uniform/mul (92.16k/92.16k flops)
block_13_project/kernel/Initializer/random_uniform/sub (1/1 flops)
Logits/kernel/Initializer/random_uniform (62.72k/125.44k flops)
Logits/kernel/Initializer/random_uniform/mul (62.72k/62.72k flops)
Logits/kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_project/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_12_project/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_12_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_11_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_11_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_12_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_12_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_project/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_11_project/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_11_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_expand/kernel/Initializer/random_uniform (55.30k/110.59k flops)
block_13_expand/kernel/Initializer/random_uniform/mul (55.30k/55.30k flops)
block_13_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_project/kernel/Initializer/random_uniform (36.86k/73.73k flops)
block_10_project/kernel/Initializer/random_uniform/mul (36.86k/36.86k flops)
block_10_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_9_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_9_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_8_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_8_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_10_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_10_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_7_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_7_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_expand/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_7_expand/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_7_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_8_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_8_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_project/kernel/Initializer/random_uniform (24.58k/49.15k flops)
block_9_project/kernel/Initializer/random_uniform/mul (24.58k/24.58k flops)
block_9_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_project/kernel/Initializer/random_uniform (12.29k/24.58k flops)
block_6_project/kernel/Initializer/random_uniform/mul (12.29k/12.29k flops)
block_6_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_16_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_15_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform (8.64k/17.28k flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform/mul (8.64k/8.64k flops)
block_14_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_5_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_5_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_project/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_5_project/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_5_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_project/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_4_project/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_4_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_6_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_6_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_expand/kernel/Initializer/random_uniform (6.14k/12.29k flops)
block_4_expand/kernel/Initializer/random_uniform/mul (6.14k/6.14k flops)
block_4_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_11_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_13_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform (5.18k/10.37k flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform/mul (5.18k/5.18k flops)
block_12_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_project/kernel/Initializer/random_uniform (4.61k/9.22k flops)
block_3_project/kernel/Initializer/random_uniform/mul (4.61k/4.61k flops)
block_3_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_expand/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_3_expand/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_3_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_project/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_2_project/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_2_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_expand/kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_2_expand/kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_2_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_10_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_7_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_8_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform (3.46k/6.91k flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform/mul (3.46k/3.46k flops)
block_9_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_project/kernel/Initializer/random_uniform (2.30k/4.61k flops)
block_1_project/kernel/Initializer/random_uniform/mul (2.30k/2.30k flops)
block_1_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_4_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_6_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform (1.73k/3.46k flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.73k/1.73k flops)
block_5_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_expand/kernel/Initializer/random_uniform (1.54k/3.07k flops)
block_1_expand/kernel/Initializer/random_uniform/mul (1.54k/1.54k flops)
block_1_expand/kernel/Initializer/random_uniform/sub (1/1 flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform (1.30k/2.59k flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.30k/1.30k flops)
block_2_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform (1.30k/2.59k flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform/mul (1.30k/1.30k flops)
block_3_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform (864/1.73k flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform/mul (864/864 flops)
block_1_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg/mul (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg/sub_1 (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg_1/mul (1.28k/1.28k flops)
Conv_1_bn/AssignMovingAvg_1/sub_1 (1.28k/1.28k flops)
expanded_conv_project/kernel/Initializer/random_uniform (512/1.02k flops)
expanded_conv_project/kernel/Initializer/random_uniform/mul (512/512 flops)
expanded_conv_project/kernel/Initializer/random_uniform/sub (1/1 flops)
block_14_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_15_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_15_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_15_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_15_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg/mul (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_14_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_16_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_16_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_16_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_16_depthwise_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_16_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_15_expand_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
block_14_expand_BN/AssignMovingAvg_1/mul (960/960 flops)
block_14_expand_BN/AssignMovingAvg/sub_1 (960/960 flops)
block_14_expand_BN/AssignMovingAvg/mul (960/960 flops)
block_14_depthwise_BN/AssignMovingAvg_1/sub_1 (960/960 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform (288/577 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform/mul (288/288 flops)
expanded_conv_depthwise/depthwise_kernel/Initializer/random_uniform/sub (1/1 flops)
Conv1/kernel/Initializer/random_uniform (288/577 flops)
Conv1/kernel/Initializer/random_uniform/mul (288/288 flops)
Conv1/kernel/Initializer/random_uniform/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_13_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_13_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_13_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_11_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_11_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_11_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_11_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_11_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_12_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_12_expand_BN/AssignMovingAvg/mul (576/576 flops)
block_12_expand_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_12_expand_BN/AssignMovingAvg_1/mul (576/576 flops)
block_12_expand_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg/mul (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg/sub_1 (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg_1/mul (576/576 flops)
block_13_depthwise_BN/AssignMovingAvg_1/sub_1 (576/576 flops)
block_7_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_10_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_9_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_9_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_9_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_8_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_10_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_10_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_8_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_10_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_7_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_10_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_7_expand_BN/AssignMovingAvg_1/mul (384/384 flops)
block_7_expand_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg/mul (384/384 flops)
block_9_depthwise_BN/AssignMovingAvg_1/mul (384/384 flops)
block_7_expand_BN/AssignMovingAvg/sub_1 (384/384 flops)
block_8_depthwise_BN/AssignMovingAvg_1/sub_1 (384/384 flops)
block_7_expand_BN/AssignMovingAvg/mul (384/384 flops)
block_16_project_BN/AssignMovingAvg/mul (320/320 flops)
block_16_project_BN/AssignMovingAvg/sub_1 (320/320 flops)
block_16_project_BN/AssignMovingAvg_1/mul (320/320 flops)
block_16_project_BN/AssignMovingAvg_1/sub_1 (320/320 flops)
block_4_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_4_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_4_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_6_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_6_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_4_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_6_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_4_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_6_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg/mul (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_5_expand_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_5_expand_BN/AssignMovingAvg_1/mul (192/192 flops)
block_5_expand_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_5_expand_BN/AssignMovingAvg/mul (192/192 flops)
block_4_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg_1/sub_1 (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_6_depthwise_BN/AssignMovingAvg_1/mul (192/192 flops)
block_5_depthwise_BN/AssignMovingAvg/sub_1 (192/192 flops)
block_14_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_14_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_14_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_15_project_BN/AssignMovingAvg/mul (160/160 flops)
block_15_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_15_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_15_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_13_project_BN/AssignMovingAvg_1/sub_1 (160/160 flops)
block_13_project_BN/AssignMovingAvg/mul (160/160 flops)
block_13_project_BN/AssignMovingAvg/sub_1 (160/160 flops)
block_13_project_BN/AssignMovingAvg_1/mul (160/160 flops)
block_14_project_BN/AssignMovingAvg/mul (160/160 flops)
block_3_depthwise_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg_1/mul (144/144 flops)
block_2_expand_BN/AssignMovingAvg_1/mul (144/144 flops)
block_3_expand_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_2_expand_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_3_expand_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_3_expand_BN/AssignMovingAvg/mul (144/144 flops)
block_2_expand_BN/AssignMovingAvg/mul (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg/mul (144/144 flops)
block_2_expand_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_3_expand_BN/AssignMovingAvg_1/mul (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg/mul (144/144 flops)
block_2_depthwise_BN/AssignMovingAvg/sub_1 (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg_1/mul (144/144 flops)
block_3_depthwise_BN/AssignMovingAvg_1/sub_1 (144/144 flops)
block_1_depthwise_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_10_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_10_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_11_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg/mul (96/96 flops)
block_12_project_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg/mul (96/96 flops)
block_12_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_11_project_BN/AssignMovingAvg_1/mul (96/96 flops)
block_11_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_12_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_10_project_BN/AssignMovingAvg/mul (96/96 flops)
block_1_expand_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_1_expand_BN/AssignMovingAvg_1/mul (96/96 flops)
block_1_expand_BN/AssignMovingAvg/sub_1 (96/96 flops)
block_1_expand_BN/AssignMovingAvg/mul (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_10_project_BN/AssignMovingAvg_1/sub_1 (96/96 flops)
block_11_project_BN/AssignMovingAvg/mul (96/96 flops)
block_1_depthwise_BN/AssignMovingAvg_1/mul (96/96 flops)
block_6_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_7_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_6_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_6_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_6_project_BN/AssignMovingAvg/mul (64/64 flops)
block_7_project_BN/AssignMovingAvg/mul (64/64 flops)
block_8_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_9_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_8_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_9_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_9_project_BN/AssignMovingAvg/mul (64/64 flops)
block_8_project_BN/AssignMovingAvg/sub_1 (64/64 flops)
block_8_project_BN/AssignMovingAvg/mul (64/64 flops)
block_9_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_7_project_BN/AssignMovingAvg_1/sub_1 (64/64 flops)
block_7_project_BN/AssignMovingAvg_1/mul (64/64 flops)
block_3_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
block_3_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
bn_Conv1/AssignMovingAvg_1/mul (32/32 flops)
bn_Conv1/AssignMovingAvg/sub_1 (32/32 flops)
block_5_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_4_project_BN/AssignMovingAvg/mul (32/32 flops)
bn_Conv1/AssignMovingAvg/mul (32/32 flops)
block_4_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
block_4_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg/mul (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_3_project_BN/AssignMovingAvg/mul (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/mul (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/sub_1 (32/32 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/mul (32/32 flops)
block_3_project_BN/AssignMovingAvg_1/mul (32/32 flops)
block_5_project_BN/AssignMovingAvg/sub_1 (32/32 flops)
bn_Conv1/AssignMovingAvg_1/sub_1 (32/32 flops)
block_4_project_BN/AssignMovingAvg_1/sub_1 (32/32 flops)
block_2_project_BN/AssignMovingAvg/mul (24/24 flops)
block_1_project_BN/AssignMovingAvg_1/mul (24/24 flops)
block_2_project_BN/AssignMovingAvg/sub_1 (24/24 flops)
block_2_project_BN/AssignMovingAvg_1/mul (24/24 flops)
block_1_project_BN/AssignMovingAvg_1/sub_1 (24/24 flops)
block_2_project_BN/AssignMovingAvg_1/sub_1 (24/24 flops)
block_1_project_BN/AssignMovingAvg/sub_1 (24/24 flops)
block_1_project_BN/AssignMovingAvg/mul (24/24 flops)
expanded_conv_project_BN/AssignMovingAvg/mul (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg/sub_1 (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg_1/mul (16/16 flops)
expanded_conv_project_BN/AssignMovingAvg_1/sub_1 (16/16 flops)
block_10_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_10_project_BN/AssignMovingAvg/sub (1/1 flops)
block_2_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_2_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_2_project_BN/AssignMovingAvg/sub (1/1 flops)
block_2_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_2_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_3_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_3_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_3_project_BN/AssignMovingAvg/sub (1/1 flops)
block_3_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_14_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_project_BN/AssignMovingAvg/sub (1/1 flops)
block_15_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_6_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_6_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_15_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_14_project_BN/AssignMovingAvg/sub (1/1 flops)
block_14_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_14_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_14_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_14_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_11_project_BN/AssignMovingAvg/sub (1/1 flops)
block_13_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_13_project_BN/AssignMovingAvg/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_13_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_13_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_project_BN/AssignMovingAvg/sub (1/1 flops)
block_15_project_BN/AssignMovingAvg/sub (1/1 flops)
block_15_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_7_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_13_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_16_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_8_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_12_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_16_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_8_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_6_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_12_project_BN/AssignMovingAvg/sub (1/1 flops)
block_16_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_8_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_12_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_8_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_12_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_16_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_8_project_BN/AssignMovingAvg/sub (1/1 flops)
block_5_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_5_project_BN/AssignMovingAvg/sub (1/1 flops)
block_8_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_16_project_BN/AssignMovingAvg/sub (1/1 flops)
block_16_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_9_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_5_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_9_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_9_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_11_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_11_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_9_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_5_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_10_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_5_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_9_project_BN/AssignMovingAvg/sub (1/1 flops)
block_1_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_4_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_9_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_5_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_4_project_BN/AssignMovingAvg/sub (1/1 flops)
bn_Conv1/AssignMovingAvg/sub (1/1 flops)
block_1_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_expand_BN/AssignMovingAvg_1/sub (1/1 flops)
bn_Conv1/AssignMovingAvg_1/sub (1/1 flops)
block_4_expand_BN/AssignMovingAvg/sub (1/1 flops)
block_1_project_BN/AssignMovingAvg/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg/sub (1/1 flops)
expanded_conv_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_4_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_1_project_BN/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
block_4_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_2_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
Conv_1_bn/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_project_BN/AssignMovingAvg/sub (1/1 flops)
block_10_depthwise_BN/AssignMovingAvg/sub (1/1 flops)
block_10_depthwise_BN/AssignMovingAvg_1/sub (1/1 flops)
expanded_conv_project_BN/AssignMovingAvg_1/sub (1/1 flops)
block_10_expand_BN/AssignMovingAvg/sub (1/1 flops)
loss/Logits_loss/softmax_cross_entropy_with_logits/Sub (1/1 flops)
loss/Logits_loss/softmax_cross_entropy_with_logits/Sub_1 (1/1 flops)
loss/Logits_loss/softmax_cross_entropy_with_logits/Sub_2 (1/1 flops)
loss/mul (1/1 flops)
======================End of Report==========================
62 ops no flops stats due to incomplete shapes.
Incomplete shape.
Incomplete shape.
Incomplete shape.
=========================Options=============================
-max_depth 10000
-min_bytes 0
-min_peak_bytes 0
-min_residual_bytes 0
-min_output_bytes 0
-min_micros 0
-min_accelerator_micros 0
-min_cpu_micros 0
-min_params 0
-min_float_ops 0
-min_occurrence 0
-step -1
-order_by name
-account_type_regexes _trainable_variables
-start_name_regexes .*
-trim_name_regexes
-show_name_regexes .*
-hide_name_regexes
-account_displayed_op_only true
-select params
-output stdout:
==================Model Analysis Report======================
Incomplete shape.
Doc:
scope: The nodes in the model graph are organized by their names, which is hierarchical like filesystem.
param: Number of parameters (in the Variable).
Profile:
node name | # parameters
_TFProfRoot (--/2.29m params)
Conv1 (--/288 params)
Conv1/kernel (3x3x1x32, 288/288 params)
Conv_1 (--/409.60k params)
Conv_1/kernel (1x1x320x1280, 409.60k/409.60k params)
Conv_1_bn (--/2.56k params)
Conv_1_bn/beta (1280, 1.28k/1.28k params)
Conv_1_bn/gamma (1280, 1.28k/1.28k params)
Logits (--/62.77k params)
Logits/bias (49, 49/49 params)
Logits/kernel (1280x49, 62.72k/62.72k params)
block_10_depthwise (--/3.46k params)
block_10_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_10_depthwise_BN (--/768 params)
block_10_depthwise_BN/beta (384, 384/384 params)
block_10_depthwise_BN/gamma (384, 384/384 params)
block_10_expand (--/24.58k params)
block_10_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_10_expand_BN (--/768 params)
block_10_expand_BN/beta (384, 384/384 params)
block_10_expand_BN/gamma (384, 384/384 params)
block_10_project (--/36.86k params)
block_10_project/kernel (1x1x384x96, 36.86k/36.86k params)
block_10_project_BN (--/192 params)
block_10_project_BN/beta (96, 96/96 params)
block_10_project_BN/gamma (96, 96/96 params)
block_11_depthwise (--/5.18k params)
block_11_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_11_depthwise_BN (--/1.15k params)
block_11_depthwise_BN/beta (576, 576/576 params)
block_11_depthwise_BN/gamma (576, 576/576 params)
block_11_expand (--/55.30k params)
block_11_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_11_expand_BN (--/1.15k params)
block_11_expand_BN/beta (576, 576/576 params)
block_11_expand_BN/gamma (576, 576/576 params)
block_11_project (--/55.30k params)
block_11_project/kernel (1x1x576x96, 55.30k/55.30k params)
block_11_project_BN (--/192 params)
block_11_project_BN/beta (96, 96/96 params)
block_11_project_BN/gamma (96, 96/96 params)
block_12_depthwise (--/5.18k params)
block_12_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_12_depthwise_BN (--/1.15k params)
block_12_depthwise_BN/beta (576, 576/576 params)
block_12_depthwise_BN/gamma (576, 576/576 params)
block_12_expand (--/55.30k params)
block_12_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_12_expand_BN (--/1.15k params)
block_12_expand_BN/beta (576, 576/576 params)
block_12_expand_BN/gamma (576, 576/576 params)
block_12_project (--/55.30k params)
block_12_project/kernel (1x1x576x96, 55.30k/55.30k params)
block_12_project_BN (--/192 params)
block_12_project_BN/beta (96, 96/96 params)
block_12_project_BN/gamma (96, 96/96 params)
block_13_depthwise (--/5.18k params)
block_13_depthwise/depthwise_kernel (3x3x576x1, 5.18k/5.18k params)
block_13_depthwise_BN (--/1.15k params)
block_13_depthwise_BN/beta (576, 576/576 params)
block_13_depthwise_BN/gamma (576, 576/576 params)
block_13_expand (--/55.30k params)
block_13_expand/kernel (1x1x96x576, 55.30k/55.30k params)
block_13_expand_BN (--/1.15k params)
block_13_expand_BN/beta (576, 576/576 params)
block_13_expand_BN/gamma (576, 576/576 params)
block_13_project (--/92.16k params)
block_13_project/kernel (1x1x576x160, 92.16k/92.16k params)
block_13_project_BN (--/320 params)
block_13_project_BN/beta (160, 160/160 params)
block_13_project_BN/gamma (160, 160/160 params)
block_14_depthwise (--/8.64k params)
block_14_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_14_depthwise_BN (--/1.92k params)
block_14_depthwise_BN/beta (960, 960/960 params)
block_14_depthwise_BN/gamma (960, 960/960 params)
block_14_expand (--/153.60k params)
block_14_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_14_expand_BN (--/1.92k params)
block_14_expand_BN/beta (960, 960/960 params)
block_14_expand_BN/gamma (960, 960/960 params)
block_14_project (--/153.60k params)
block_14_project/kernel (1x1x960x160, 153.60k/153.60k params)
block_14_project_BN (--/320 params)
block_14_project_BN/beta (160, 160/160 params)
block_14_project_BN/gamma (160, 160/160 params)
block_15_depthwise (--/8.64k params)
block_15_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_15_depthwise_BN (--/1.92k params)
block_15_depthwise_BN/beta (960, 960/960 params)
block_15_depthwise_BN/gamma (960, 960/960 params)
block_15_expand (--/153.60k params)
block_15_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_15_expand_BN (--/1.92k params)
block_15_expand_BN/beta (960, 960/960 params)
block_15_expand_BN/gamma (960, 960/960 params)
block_15_project (--/153.60k params)
block_15_project/kernel (1x1x960x160, 153.60k/153.60k params)
block_15_project_BN (--/320 params)
block_15_project_BN/beta (160, 160/160 params)
block_15_project_BN/gamma (160, 160/160 params)
block_16_depthwise (--/8.64k params)
block_16_depthwise/depthwise_kernel (3x3x960x1, 8.64k/8.64k params)
block_16_depthwise_BN (--/1.92k params)
block_16_depthwise_BN/beta (960, 960/960 params)
block_16_depthwise_BN/gamma (960, 960/960 params)
block_16_expand (--/153.60k params)
block_16_expand/kernel (1x1x160x960, 153.60k/153.60k params)
block_16_expand_BN (--/1.92k params)
block_16_expand_BN/beta (960, 960/960 params)
block_16_expand_BN/gamma (960, 960/960 params)
block_16_project (--/307.20k params)
block_16_project/kernel (1x1x960x320, 307.20k/307.20k params)
block_16_project_BN (--/640 params)
block_16_project_BN/beta (320, 320/320 params)
block_16_project_BN/gamma (320, 320/320 params)
block_1_depthwise (--/864 params)
block_1_depthwise/depthwise_kernel (3x3x96x1, 864/864 params)
block_1_depthwise_BN (--/192 params)
block_1_depthwise_BN/beta (96, 96/96 params)
block_1_depthwise_BN/gamma (96, 96/96 params)
block_1_expand (--/1.54k params)
block_1_expand/kernel (1x1x16x96, 1.54k/1.54k params)
block_1_expand_BN (--/192 params)
block_1_expand_BN/beta (96, 96/96 params)
block_1_expand_BN/gamma (96, 96/96 params)
block_1_project (--/2.30k params)
block_1_project/kernel (1x1x96x24, 2.30k/2.30k params)
block_1_project_BN (--/48 params)
block_1_project_BN/beta (24, 24/24 params)
block_1_project_BN/gamma (24, 24/24 params)
block_2_depthwise (--/1.30k params)
block_2_depthwise/depthwise_kernel (3x3x144x1, 1.30k/1.30k params)
block_2_depthwise_BN (--/288 params)
block_2_depthwise_BN/beta (144, 144/144 params)
block_2_depthwise_BN/gamma (144, 144/144 params)
block_2_expand (--/3.46k params)
block_2_expand/kernel (1x1x24x144, 3.46k/3.46k params)
block_2_expand_BN (--/288 params)
block_2_expand_BN/beta (144, 144/144 params)
block_2_expand_BN/gamma (144, 144/144 params)
block_2_project (--/3.46k params)
block_2_project/kernel (1x1x144x24, 3.46k/3.46k params)
block_2_project_BN (--/48 params)
block_2_project_BN/beta (24, 24/24 params)
block_2_project_BN/gamma (24, 24/24 params)
block_3_depthwise (--/1.30k params)
block_3_depthwise/depthwise_kernel (3x3x144x1, 1.30k/1.30k params)
block_3_depthwise_BN (--/288 params)
block_3_depthwise_BN/beta (144, 144/144 params)
block_3_depthwise_BN/gamma (144, 144/144 params)
block_3_expand (--/3.46k params)
block_3_expand/kernel (1x1x24x144, 3.46k/3.46k params)
block_3_expand_BN (--/288 params)
block_3_expand_BN/beta (144, 144/144 params)
block_3_expand_BN/gamma (144, 144/144 params)
block_3_project (--/4.61k params)
block_3_project/kernel (1x1x144x32, 4.61k/4.61k params)
block_3_project_BN (--/64 params)
block_3_project_BN/beta (32, 32/32 params)
block_3_project_BN/gamma (32, 32/32 params)
block_4_depthwise (--/1.73k params)
block_4_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_4_depthwise_BN (--/384 params)
block_4_depthwise_BN/beta (192, 192/192 params)
block_4_depthwise_BN/gamma (192, 192/192 params)
block_4_expand (--/6.14k params)
block_4_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_4_expand_BN (--/384 params)
block_4_expand_BN/beta (192, 192/192 params)
block_4_expand_BN/gamma (192, 192/192 params)
block_4_project (--/6.14k params)
block_4_project/kernel (1x1x192x32, 6.14k/6.14k params)
block_4_project_BN (--/64 params)
block_4_project_BN/beta (32, 32/32 params)
block_4_project_BN/gamma (32, 32/32 params)
block_5_depthwise (--/1.73k params)
block_5_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_5_depthwise_BN (--/384 params)
block_5_depthwise_BN/beta (192, 192/192 params)
block_5_depthwise_BN/gamma (192, 192/192 params)
block_5_expand (--/6.14k params)
block_5_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_5_expand_BN (--/384 params)
block_5_expand_BN/beta (192, 192/192 params)
block_5_expand_BN/gamma (192, 192/192 params)
block_5_project (--/6.14k params)
block_5_project/kernel (1x1x192x32, 6.14k/6.14k params)
block_5_project_BN (--/64 params)
block_5_project_BN/beta (32, 32/32 params)
block_5_project_BN/gamma (32, 32/32 params)
block_6_depthwise (--/1.73k params)
block_6_depthwise/depthwise_kernel (3x3x192x1, 1.73k/1.73k params)
block_6_depthwise_BN (--/384 params)
block_6_depthwise_BN/beta (192, 192/192 params)
block_6_depthwise_BN/gamma (192, 192/192 params)
block_6_expand (--/6.14k params)
block_6_expand/kernel (1x1x32x192, 6.14k/6.14k params)
block_6_expand_BN (--/384 params)
block_6_expand_BN/beta (192, 192/192 params)
block_6_expand_BN/gamma (192, 192/192 params)
block_6_project (--/12.29k params)
block_6_project/kernel (1x1x192x64, 12.29k/12.29k params)
block_6_project_BN (--/128 params)
block_6_project_BN/beta (64, 64/64 params)
block_6_project_BN/gamma (64, 64/64 params)
block_7_depthwise (--/3.46k params)
block_7_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_7_depthwise_BN (--/768 params)
block_7_depthwise_BN/beta (384, 384/384 params)
block_7_depthwise_BN/gamma (384, 384/384 params)
block_7_expand (--/24.58k params)
block_7_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_7_expand_BN (--/768 params)
block_7_expand_BN/beta (384, 384/384 params)
block_7_expand_BN/gamma (384, 384/384 params)
block_7_project (--/24.58k params)
block_7_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_7_project_BN (--/128 params)
block_7_project_BN/beta (64, 64/64 params)
block_7_project_BN/gamma (64, 64/64 params)
block_8_depthwise (--/3.46k params)
block_8_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_8_depthwise_BN (--/768 params)
block_8_depthwise_BN/beta (384, 384/384 params)
block_8_depthwise_BN/gamma (384, 384/384 params)
block_8_expand (--/24.58k params)
block_8_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_8_expand_BN (--/768 params)
block_8_expand_BN/beta (384, 384/384 params)
block_8_expand_BN/gamma (384, 384/384 params)
block_8_project (--/24.58k params)
block_8_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_8_project_BN (--/128 params)
block_8_project_BN/beta (64, 64/64 params)
block_8_project_BN/gamma (64, 64/64 params)
block_9_depthwise (--/3.46k params)
block_9_depthwise/depthwise_kernel (3x3x384x1, 3.46k/3.46k params)
block_9_depthwise_BN (--/768 params)
block_9_depthwise_BN/beta (384, 384/384 params)
block_9_depthwise_BN/gamma (384, 384/384 params)
block_9_expand (--/24.58k params)
block_9_expand/kernel (1x1x64x384, 24.58k/24.58k params)
block_9_expand_BN (--/768 params)
block_9_expand_BN/beta (384, 384/384 params)
block_9_expand_BN/gamma (384, 384/384 params)
block_9_project (--/24.58k params)
block_9_project/kernel (1x1x384x64, 24.58k/24.58k params)
block_9_project_BN (--/128 params)
block_9_project_BN/beta (64, 64/64 params)
block_9_project_BN/gamma (64, 64/64 params)
bn_Conv1 (--/64 params)
bn_Conv1/beta (32, 32/32 params)
bn_Conv1/gamma (32, 32/32 params)
expanded_conv_depthwise (--/288 params)
expanded_conv_depthwise/depthwise_kernel (3x3x32x1, 288/288 params)
expanded_conv_depthwise_BN (--/64 params)
expanded_conv_depthwise_BN/beta (32, 32/32 params)
expanded_conv_depthwise_BN/gamma (32, 32/32 params)
expanded_conv_project (--/512 params)
expanded_conv_project/kernel (1x1x32x16, 512/512 params)
expanded_conv_project_BN (--/32 params)
expanded_conv_project_BN/beta (16, 16/16 params)
expanded_conv_project_BN/gamma (16, 16/16 params)
======================End of Report==========================
FLOPs: 4,572,193; Trainable params: 2,286,065
WARNING:tensorflow:From E:/master_ImRecognition/main.py:99: The name tf.keras.backend.get_session is deprecated. Please use tf.compat.v1.keras.backend.get_session instead.
[0418 21:45:19] From E:/master_ImRecognition/main.py:99: The name tf.keras.backend.get_session is deprecated. Please use tf.compat.v1.keras.backend.get_session instead.
2021-04-18 21:45:19.082183: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1261] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-04-18 21:45:19.082353: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267]
2021-04-18 21:45:19.082452: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
2021-04-18 21:45:19.257091: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:196] None of the MLIR optimization passes are enabled (registered 0 passes)
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
=========================Options=============================
-max_depth 10000
-min_bytes 0
-min_peak_bytes 0
-min_residual_bytes 0
-min_output_bytes 0
-min_micros 0
-min_accelerator_micros 0
-min_cpu_micros 0
-min_params 0
-min_float_ops 1
-min_occurrence 0
-step -1
-order_by float_ops
-account_type_regexes .*
-start_name_regexes .*
-trim_name_regexes
-show_name_regexes .*
-hide_name_regexes
-account_displayed_op_only true
-select float_ops
-output stdout:
==================Model Analysis Report======================
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Incomplete shape.
Doc:
op: The nodes are operation kernel type, such as MatMul, Conv2D. Graph nodes belonging to the same type are aggregated together.
flops: Number of float operations. Note: Please read the implementation for the math behind it.
Profile:
node name | # float_ops
Mul 2.29m float_ops (100.00%, 50.00%)
Add 2.25m float_ops (50.00%, 49.25%)
Sub 34.27k float_ops (0.75%, 0.75%)
======================End of Report==========================
Model: "mobilenetv2_1.00_200"
方法二:
def get_flops(model):
run_meta = tf.RunMetadata()
opts = tf.profiler.ProfileOptionBuilder.float_operation()
flops = tf.profiler.profile(graph=K.get_session().graph,
run_meta=run_meta,
cmd='op',
options=opts)
return flops.total_float_ops
结果与上述方法一基本相同,不同TensorFlow版本的代码差异。
方法三:
使用net_flops()函数的结果,与MobieNetV2论文结果较接近,但仍存在一定的误差:
Layer Name | Input Shape | Output Shape | Kernel Size | Filters | Strides | FLOPS
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
input_1 | [224, 224, 3] | [224, 224, 3] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv1 | [224, 224, 3] | [112, 112, 32] | (3, 3) | 32 | (2, 2) | 21676032.0000
bn_Conv1 | [112, 112, 32] | [112, 112, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv1_relu | [112, 112, 32] | [112, 112, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_depthwise | [112, 112, 32] | [112, 112, 32] | (3, 3) | 32 | (1, 1) | 7225344.0000
expanded_conv_depthwise_BN | [112, 112, 32] | [112, 112, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_depthwise_relu | [112, 112, 32] | [112, 112, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_project | [112, 112, 32] | [112, 112, 16] | (1, 1) | 16 | (1, 1) | 12845056.0000
expanded_conv_project_BN | [112, 112, 16] | [112, 112, 16] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_expand | [112, 112, 16] | [112, 112, 96] | (1, 1) | 96 | (1, 1) | 38535168.0000
block_1_expand_BN | [112, 112, 96] | [112, 112, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_expand_relu | [112, 112, 96] | [112, 112, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_depthwise | [113, 113, 96] | [56, 56, 96] | (3, 3) | 96 | (2, 2) | 5516208.0000
block_1_depthwise_BN | [56, 56, 96] | [56, 56, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_depthwise_relu | [56, 56, 96] | [56, 56, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_project | [56, 56, 96] | [56, 56, 24] | (1, 1) | 24 | (1, 1) | 14450688.0000
block_1_project_BN | [56, 56, 24] | [56, 56, 24] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_expand | [56, 56, 24] | [56, 56, 144] | (1, 1) | 144 | (1, 1) | 21676032.0000
block_2_expand_BN | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_expand_relu | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_depthwise | [56, 56, 144] | [56, 56, 144] | (3, 3) | 144 | (1, 1) | 8128512.0000
block_2_depthwise_BN | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_depthwise_relu | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_project | [56, 56, 144] | [56, 56, 24] | (1, 1) | 24 | (1, 1) | 21676032.0000
block_2_project_BN | [56, 56, 24] | [56, 56, 24] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_add | [56, 56, 24, 2] | [56, 56, 24] | [0, 0] | [0, 0] | [1, 1] | 75264.0000
block_3_expand | [56, 56, 24] | [56, 56, 144] | (1, 1) | 144 | (1, 1) | 21676032.0000
block_3_expand_BN | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_expand_relu | [56, 56, 144] | [56, 56, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_depthwise | [57, 57, 144] | [28, 28, 144] | (3, 3) | 144 | (2, 2) | 2105352.0000
block_3_depthwise_BN | [28, 28, 144] | [28, 28, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_depthwise_relu | [28, 28, 144] | [28, 28, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_project | [28, 28, 144] | [28, 28, 32] | (1, 1) | 32 | (1, 1) | 7225344.0000
block_3_project_BN | [28, 28, 32] | [28, 28, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_expand | [28, 28, 32] | [28, 28, 192] | (1, 1) | 192 | (1, 1) | 9633792.0000
block_4_expand_BN | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_expand_relu | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_depthwise | [28, 28, 192] | [28, 28, 192] | (3, 3) | 192 | (1, 1) | 2709504.0000
block_4_depthwise_BN | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_depthwise_relu | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_project | [28, 28, 192] | [28, 28, 32] | (1, 1) | 32 | (1, 1) | 9633792.0000
block_4_project_BN | [28, 28, 32] | [28, 28, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_add | [28, 28, 32, 2] | [28, 28, 32] | [0, 0] | [0, 0] | [1, 1] | 25088.0000
block_5_expand | [28, 28, 32] | [28, 28, 192] | (1, 1) | 192 | (1, 1) | 9633792.0000
block_5_expand_BN | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_expand_relu | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_depthwise | [28, 28, 192] | [28, 28, 192] | (3, 3) | 192 | (1, 1) | 2709504.0000
block_5_depthwise_BN | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_depthwise_relu | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_project | [28, 28, 192] | [28, 28, 32] | (1, 1) | 32 | (1, 1) | 9633792.0000
block_5_project_BN | [28, 28, 32] | [28, 28, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_add | [28, 28, 32, 2] | [28, 28, 32] | [0, 0] | [0, 0] | [1, 1] | 25088.0000
block_6_expand | [28, 28, 32] | [28, 28, 192] | (1, 1) | 192 | (1, 1) | 9633792.0000
block_6_expand_BN | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_expand_relu | [28, 28, 192] | [28, 28, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_depthwise | [29, 29, 192] | [14, 14, 192] | (3, 3) | 192 | (2, 2) | 726624.0000
block_6_depthwise_BN | [14, 14, 192] | [14, 14, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_depthwise_relu | [14, 14, 192] | [14, 14, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_project | [14, 14, 192] | [14, 14, 64] | (1, 1) | 64 | (1, 1) | 4816896.0000
block_6_project_BN | [14, 14, 64] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_expand | [14, 14, 64] | [14, 14, 384] | (1, 1) | 384 | (1, 1) | 9633792.0000
block_7_expand_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_expand_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_depthwise | [14, 14, 384] | [14, 14, 384] | (3, 3) | 384 | (1, 1) | 1354752.0000
block_7_depthwise_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_depthwise_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_project | [14, 14, 384] | [14, 14, 64] | (1, 1) | 64 | (1, 1) | 9633792.0000
block_7_project_BN | [14, 14, 64] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_add | [14, 14, 64, 2] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 12544.0000
block_8_expand | [14, 14, 64] | [14, 14, 384] | (1, 1) | 384 | (1, 1) | 9633792.0000
block_8_expand_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_expand_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_depthwise | [14, 14, 384] | [14, 14, 384] | (3, 3) | 384 | (1, 1) | 1354752.0000
block_8_depthwise_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_depthwise_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_project | [14, 14, 384] | [14, 14, 64] | (1, 1) | 64 | (1, 1) | 9633792.0000
block_8_project_BN | [14, 14, 64] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_add | [14, 14, 64, 2] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 12544.0000
block_9_expand | [14, 14, 64] | [14, 14, 384] | (1, 1) | 384 | (1, 1) | 9633792.0000
block_9_expand_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_expand_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_depthwise | [14, 14, 384] | [14, 14, 384] | (3, 3) | 384 | (1, 1) | 1354752.0000
block_9_depthwise_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_depthwise_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_project | [14, 14, 384] | [14, 14, 64] | (1, 1) | 64 | (1, 1) | 9633792.0000
block_9_project_BN | [14, 14, 64] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_add | [14, 14, 64, 2] | [14, 14, 64] | [0, 0] | [0, 0] | [1, 1] | 12544.0000
block_10_expand | [14, 14, 64] | [14, 14, 384] | (1, 1) | 384 | (1, 1) | 9633792.0000
block_10_expand_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_expand_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_depthwise | [14, 14, 384] | [14, 14, 384] | (3, 3) | 384 | (1, 1) | 1354752.0000
block_10_depthwise_BN | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_depthwise_relu | [14, 14, 384] | [14, 14, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_project | [14, 14, 384] | [14, 14, 96] | (1, 1) | 96 | (1, 1) | 14450688.0000
block_10_project_BN | [14, 14, 96] | [14, 14, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_expand | [14, 14, 96] | [14, 14, 576] | (1, 1) | 576 | (1, 1) | 21676032.0000
block_11_expand_BN | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_expand_relu | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_depthwise | [14, 14, 576] | [14, 14, 576] | (3, 3) | 576 | (1, 1) | 2032128.0000
block_11_depthwise_BN | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_depthwise_relu | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_project | [14, 14, 576] | [14, 14, 96] | (1, 1) | 96 | (1, 1) | 21676032.0000
block_11_project_BN | [14, 14, 96] | [14, 14, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_add | [14, 14, 96, 2] | [14, 14, 96] | [0, 0] | [0, 0] | [1, 1] | 18816.0000
block_12_expand | [14, 14, 96] | [14, 14, 576] | (1, 1) | 576 | (1, 1) | 21676032.0000
block_12_expand_BN | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_expand_relu | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_depthwise | [14, 14, 576] | [14, 14, 576] | (3, 3) | 576 | (1, 1) | 2032128.0000
block_12_depthwise_BN | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_depthwise_relu | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_project | [14, 14, 576] | [14, 14, 96] | (1, 1) | 96 | (1, 1) | 21676032.0000
block_12_project_BN | [14, 14, 96] | [14, 14, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_add | [14, 14, 96, 2] | [14, 14, 96] | [0, 0] | [0, 0] | [1, 1] | 18816.0000
block_13_expand | [14, 14, 96] | [14, 14, 576] | (1, 1) | 576 | (1, 1) | 21676032.0000
block_13_expand_BN | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_expand_relu | [14, 14, 576] | [14, 14, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_depthwise | [15, 15, 576] | [7, 7, 576] | (3, 3) | 576 | (2, 2) | 583200.0000
block_13_depthwise_BN | [7, 7, 576] | [7, 7, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_depthwise_relu | [7, 7, 576] | [7, 7, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_project | [7, 7, 576] | [7, 7, 160] | (1, 1) | 160 | (1, 1) | 9031680.0000
block_13_project_BN | [7, 7, 160] | [7, 7, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_expand | [7, 7, 160] | [7, 7, 960] | (1, 1) | 960 | (1, 1) | 15052800.0000
block_14_expand_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_expand_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_depthwise | [7, 7, 960] | [7, 7, 960] | (3, 3) | 960 | (1, 1) | 846720.0000
block_14_depthwise_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_depthwise_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_project | [7, 7, 960] | [7, 7, 160] | (1, 1) | 160 | (1, 1) | 15052800.0000
block_14_project_BN | [7, 7, 160] | [7, 7, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_add | [7, 7, 160, 2] | [7, 7, 160] | [0, 0] | [0, 0] | [1, 1] | 7840.0000
block_15_expand | [7, 7, 160] | [7, 7, 960] | (1, 1) | 960 | (1, 1) | 15052800.0000
block_15_expand_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_expand_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_depthwise | [7, 7, 960] | [7, 7, 960] | (3, 3) | 960 | (1, 1) | 846720.0000
block_15_depthwise_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_depthwise_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_project | [7, 7, 960] | [7, 7, 160] | (1, 1) | 160 | (1, 1) | 15052800.0000
block_15_project_BN | [7, 7, 160] | [7, 7, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_add | [7, 7, 160, 2] | [7, 7, 160] | [0, 0] | [0, 0] | [1, 1] | 7840.0000
block_16_expand | [7, 7, 160] | [7, 7, 960] | (1, 1) | 960 | (1, 1) | 15052800.0000
block_16_expand_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_expand_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_depthwise | [7, 7, 960] | [7, 7, 960] | (3, 3) | 960 | (1, 1) | 846720.0000
block_16_depthwise_BN | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_depthwise_relu | [7, 7, 960] | [7, 7, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_project | [7, 7, 960] | [7, 7, 320] | (1, 1) | 320 | (1, 1) | 30105600.0000
block_16_project_BN | [7, 7, 320] | [7, 7, 320] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv_1 | [7, 7, 320] | [7, 7, 1280] | (1, 1) | 1280 | (1, 1) | 40140800.0000
Conv_1_bn | [7, 7, 1280] | [7, 7, 1280] | [0, 0] | [0, 0] | [1, 1] | 0.0000
out_relu | [7, 7, 1280] | [7, 7, 1280] | [0, 0] | [0, 0] | [1, 1] | 0.0000
global_average_pooling2d | [7, 7, 1280] | [[1280], 1, 1] | [0, 0] | [0, 0] | [1, 1] | 62720.0000
Tensor("global_average_pooling2d/Mean:0", shape=(None, 1280), dtype=float32)
predictions | 1280 | [1000] | [0, 0] | [0, 0] | [1, 1] | 2560000.0000
Total FLOPs: 602,122,488.000000
Total MACCs: 300,921,692.000000
以上是keras的application中给出的代码实现,以下是自实现MobileNet的结果测试:
2021-04-18 21:45:15.332238: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1757] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
Layer Name | Input Shape | Output Shape | Kernel Size | Filters | Strides | FLOPS
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
input_1 | [200, 12, 1] | [200, 12, 1] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv1_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv1 | [201, 13, 1] | [100, 6, 32] | (3, 3) | 32 | (2, 2) | 376272.0000
bn_Conv1 | [100, 6, 32] | [100, 6, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv1_relu | [100, 6, 32] | [100, 6, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_depthwise | [100, 6, 32] | [100, 6, 32] | (3, 3) | 32 | (1, 1) | 345600.0000
expanded_conv_depthwise_BN | [100, 6, 32] | [100, 6, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_depthwise_relu | [100, 6, 32] | [100, 6, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
expanded_conv_project | [100, 6, 32] | [100, 6, 16] | (1, 1) | 16 | (1, 1) | 614400.0000
expanded_conv_project_BN | [100, 6, 16] | [100, 6, 16] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_expand | [100, 6, 16] | [100, 6, 96] | (1, 1) | 96 | (1, 1) | 1843200.0000
block_1_expand_BN | [100, 6, 96] | [100, 6, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_expand_relu | [100, 6, 96] | [100, 6, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_depthwise | [101, 7, 96] | [50, 3, 96] | (3, 3) | 96 | (2, 2) | 305424.0000
block_1_depthwise_BN | [50, 3, 96] | [50, 3, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_depthwise_relu | [50, 3, 96] | [50, 3, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_1_project | [50, 3, 96] | [50, 3, 24] | (1, 1) | 24 | (1, 1) | 691200.0000
block_1_project_BN | [50, 3, 24] | [50, 3, 24] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_expand | [50, 3, 24] | [50, 3, 144] | (1, 1) | 144 | (1, 1) | 1036800.0000
block_2_expand_BN | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_expand_relu | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_depthwise | [50, 3, 144] | [50, 3, 144] | (3, 3) | 144 | (1, 1) | 388800.0000
block_2_depthwise_BN | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_depthwise_relu | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_project | [50, 3, 144] | [50, 3, 24] | (1, 1) | 24 | (1, 1) | 1036800.0000
block_2_project_BN | [50, 3, 24] | [50, 3, 24] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_2_add | [50, 3, 24, 2] | [50, 3, 24] | [0, 0] | [0, 0] | [1, 1] | 3600.0000
block_3_expand | [50, 3, 24] | [50, 3, 144] | (1, 1) | 144 | (1, 1) | 1036800.0000
block_3_expand_BN | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_expand_relu | [50, 3, 144] | [50, 3, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_depthwise | [51, 5, 144] | [25, 2, 144] | (3, 3) | 144 | (2, 2) | 165240.0000
block_3_depthwise_BN | [25, 2, 144] | [25, 2, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_depthwise_relu | [25, 2, 144] | [25, 2, 144] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_3_project | [25, 2, 144] | [25, 2, 32] | (1, 1) | 32 | (1, 1) | 460800.0000
block_3_project_BN | [25, 2, 32] | [25, 2, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_expand | [25, 2, 32] | [25, 2, 192] | (1, 1) | 192 | (1, 1) | 614400.0000
block_4_expand_BN | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_expand_relu | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_depthwise | [25, 2, 192] | [25, 2, 192] | (3, 3) | 192 | (1, 1) | 172800.0000
block_4_depthwise_BN | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_depthwise_relu | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_project | [25, 2, 192] | [25, 2, 32] | (1, 1) | 32 | (1, 1) | 614400.0000
block_4_project_BN | [25, 2, 32] | [25, 2, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_4_add | [25, 2, 32, 2] | [25, 2, 32] | [0, 0] | [0, 0] | [1, 1] | 1600.0000
block_5_expand | [25, 2, 32] | [25, 2, 192] | (1, 1) | 192 | (1, 1) | 614400.0000
block_5_expand_BN | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_expand_relu | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_depthwise | [25, 2, 192] | [25, 2, 192] | (3, 3) | 192 | (1, 1) | 172800.0000
block_5_depthwise_BN | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_depthwise_relu | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_project | [25, 2, 192] | [25, 2, 32] | (1, 1) | 32 | (1, 1) | 614400.0000
block_5_project_BN | [25, 2, 32] | [25, 2, 32] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_5_add | [25, 2, 32, 2] | [25, 2, 32] | [0, 0] | [0, 0] | [1, 1] | 1600.0000
block_6_expand | [25, 2, 32] | [25, 2, 192] | (1, 1) | 192 | (1, 1) | 614400.0000
block_6_expand_BN | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_expand_relu | [25, 2, 192] | [25, 2, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_depthwise | [27, 3, 192] | [13, 1, 192] | (3, 3) | 192 | (2, 2) | 69984.0000
block_6_depthwise_BN | [13, 1, 192] | [13, 1, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_depthwise_relu | [13, 1, 192] | [13, 1, 192] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_6_project | [13, 1, 192] | [13, 1, 64] | (1, 1) | 64 | (1, 1) | 319488.0000
block_6_project_BN | [13, 1, 64] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_expand | [13, 1, 64] | [13, 1, 384] | (1, 1) | 384 | (1, 1) | 638976.0000
block_7_expand_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_expand_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_depthwise | [13, 1, 384] | [13, 1, 384] | (3, 3) | 384 | (1, 1) | 89856.0000
block_7_depthwise_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_depthwise_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_project | [13, 1, 384] | [13, 1, 64] | (1, 1) | 64 | (1, 1) | 638976.0000
block_7_project_BN | [13, 1, 64] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_7_add | [13, 1, 64, 2] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 832.0000
block_8_expand | [13, 1, 64] | [13, 1, 384] | (1, 1) | 384 | (1, 1) | 638976.0000
block_8_expand_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_expand_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_depthwise | [13, 1, 384] | [13, 1, 384] | (3, 3) | 384 | (1, 1) | 89856.0000
block_8_depthwise_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_depthwise_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_project | [13, 1, 384] | [13, 1, 64] | (1, 1) | 64 | (1, 1) | 638976.0000
block_8_project_BN | [13, 1, 64] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_8_add | [13, 1, 64, 2] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 832.0000
block_9_expand | [13, 1, 64] | [13, 1, 384] | (1, 1) | 384 | (1, 1) | 638976.0000
block_9_expand_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_expand_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_depthwise | [13, 1, 384] | [13, 1, 384] | (3, 3) | 384 | (1, 1) | 89856.0000
block_9_depthwise_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_depthwise_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_project | [13, 1, 384] | [13, 1, 64] | (1, 1) | 64 | (1, 1) | 638976.0000
block_9_project_BN | [13, 1, 64] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_9_add | [13, 1, 64, 2] | [13, 1, 64] | [0, 0] | [0, 0] | [1, 1] | 832.0000
block_10_expand | [13, 1, 64] | [13, 1, 384] | (1, 1) | 384 | (1, 1) | 638976.0000
block_10_expand_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_expand_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_depthwise | [13, 1, 384] | [13, 1, 384] | (3, 3) | 384 | (1, 1) | 89856.0000
block_10_depthwise_BN | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_depthwise_relu | [13, 1, 384] | [13, 1, 384] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_10_project | [13, 1, 384] | [13, 1, 96] | (1, 1) | 96 | (1, 1) | 958464.0000
block_10_project_BN | [13, 1, 96] | [13, 1, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_expand | [13, 1, 96] | [13, 1, 576] | (1, 1) | 576 | (1, 1) | 1437696.0000
block_11_expand_BN | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_expand_relu | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_depthwise | [13, 1, 576] | [13, 1, 576] | (3, 3) | 576 | (1, 1) | 134784.0000
block_11_depthwise_BN | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_depthwise_relu | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_project | [13, 1, 576] | [13, 1, 96] | (1, 1) | 96 | (1, 1) | 1437696.0000
block_11_project_BN | [13, 1, 96] | [13, 1, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_11_add | [13, 1, 96, 2] | [13, 1, 96] | [0, 0] | [0, 0] | [1, 1] | 1248.0000
block_12_expand | [13, 1, 96] | [13, 1, 576] | (1, 1) | 576 | (1, 1) | 1437696.0000
block_12_expand_BN | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_expand_relu | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_depthwise | [13, 1, 576] | [13, 1, 576] | (3, 3) | 576 | (1, 1) | 134784.0000
block_12_depthwise_BN | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_depthwise_relu | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_project | [13, 1, 576] | [13, 1, 96] | (1, 1) | 96 | (1, 1) | 1437696.0000
block_12_project_BN | [13, 1, 96] | [13, 1, 96] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_12_add | [13, 1, 96, 2] | [13, 1, 96] | [0, 0] | [0, 0] | [1, 1] | 1248.0000
block_13_expand | [13, 1, 96] | [13, 1, 576] | (1, 1) | 576 | (1, 1) | 1437696.0000
block_13_expand_BN | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_expand_relu | [13, 1, 576] | [13, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_pad | ['', '', ''] | ['', '', ''] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_depthwise | [15, 3, 576] | [7, 1, 576] | (3, 3) | 576 | (2, 2) | 116640.0000
block_13_depthwise_BN | [7, 1, 576] | [7, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_depthwise_relu | [7, 1, 576] | [7, 1, 576] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_13_project | [7, 1, 576] | [7, 1, 160] | (1, 1) | 160 | (1, 1) | 1290240.0000
block_13_project_BN | [7, 1, 160] | [7, 1, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_expand | [7, 1, 160] | [7, 1, 960] | (1, 1) | 960 | (1, 1) | 2150400.0000
block_14_expand_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_expand_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_depthwise | [7, 1, 960] | [7, 1, 960] | (3, 3) | 960 | (1, 1) | 120960.0000
block_14_depthwise_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_depthwise_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_project | [7, 1, 960] | [7, 1, 160] | (1, 1) | 160 | (1, 1) | 2150400.0000
block_14_project_BN | [7, 1, 160] | [7, 1, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_14_add | [7, 1, 160, 2] | [7, 1, 160] | [0, 0] | [0, 0] | [1, 1] | 1120.0000
block_15_expand | [7, 1, 160] | [7, 1, 960] | (1, 1) | 960 | (1, 1) | 2150400.0000
block_15_expand_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_expand_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_depthwise | [7, 1, 960] | [7, 1, 960] | (3, 3) | 960 | (1, 1) | 120960.0000
block_15_depthwise_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_depthwise_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_project | [7, 1, 960] | [7, 1, 160] | (1, 1) | 160 | (1, 1) | 2150400.0000
block_15_project_BN | [7, 1, 160] | [7, 1, 160] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_15_add | [7, 1, 160, 2] | [7, 1, 160] | [0, 0] | [0, 0] | [1, 1] | 1120.0000
block_16_expand | [7, 1, 160] | [7, 1, 960] | (1, 1) | 960 | (1, 1) | 2150400.0000
block_16_expand_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_expand_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_depthwise | [7, 1, 960] | [7, 1, 960] | (3, 3) | 960 | (1, 1) | 120960.0000
block_16_depthwise_BN | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_depthwise_relu | [7, 1, 960] | [7, 1, 960] | [0, 0] | [0, 0] | [1, 1] | 0.0000
block_16_project | [7, 1, 960] | [7, 1, 320] | (1, 1) | 320 | (1, 1) | 4300800.0000
block_16_project_BN | [7, 1, 320] | [7, 1, 320] | [0, 0] | [0, 0] | [1, 1] | 0.0000
Conv_1 | [7, 1, 320] | [7, 1, 1280] | (1, 1) | 1280 | (1, 1) | 5734400.0000
Conv_1_bn | [7, 1, 1280] | [7, 1, 1280] | [0, 0] | [0, 0] | [1, 1] | 0.0000
out_relu | [7, 1, 1280] | [7, 1, 1280] | [0, 0] | [0, 0] | [1, 1] | 0.0000
global_average_pooling2d | [7, 1, 1280] | [[1280], 1, 1] | [0, 0] | [0, 0] | [1, 1] | 8960.0000
Tensor("global_average_pooling2d/Mean:0", shape=(None, 1280), dtype=float32)
Logits | 1280 | [49] | [0, 0] | [0, 0] | [1, 1] | 125440.0000
Total FLOPs: 48,062,568.000000
Total MACCs: 24,019,788.000000
以下是论文原文给出的数据:
版权声明:本文为jwen_原创文章,遵循 CC 4.0 BY-SA 版权协议,转载请附上原文出处链接和本声明。