Pytorch转tflite方式

目标是想把在服务器上用pytorch训练好的模型转换为可以在移动端运行的tflite模型。

最直接的思路是想把pytorch模型转换为tensorflow的模型，然后转换为tflite。但是这个转换目前没有发现比较靠谱的方法。

经过调研发现最新的tflite已经支持直接从keras模型的转换，所以可以采用keras作为中间转换的桥梁，这样就能充分利用keras高层API的便利性。

转换的基本思想就是用pytorch中的各层网络的权重取出来后直接赋值给keras网络中的对应layer层的权重。

转换为Keras模型后，再通过tf.contrib.lite.TocoConverter把模型直接转为tflite.

下面是一个例子，假设转换的是一个两层的CNN网络。

import tensorflow as tf
from tensorflow import keras
import numpy as np

import torch
from torchvision import models
import torch.nn as nn
# import torch.nn.functional as F
from torch.autograd import Variable

class PytorchNet(nn.Module):
def __init__(self):
super(PytorchNet, self).__init__()
conv1 = nn.Sequential(
 nn.Conv2d(3, 32, 3, 2),
 nn.BatchNorm2d(32),
 nn.ReLU(inplace=True),
 nn.MaxPool2d(2, 2))
conv2 = nn.Sequential(
 nn.Conv2d(32, 64, 3, 1, groups=1),
 nn.BatchNorm2d(64),
 nn.ReLU(inplace=True),
 nn.MaxPool2d(2, 2))
self.feature = nn.Sequential(conv1, conv2)
self.init_weights()

def forward(self, x):
return self.feature(x)

def init_weights(self):
for m in self.modules():
 if isinstance(m, nn.Conv2d):
 nn.init.kaiming_normal_(
  m.weight.data, mode='fan_out', nonlinearity='relu')
 if m.bias is not None:
  m.bias.data.zero_()
 if isinstance(m, nn.BatchNorm2d):
 m.weight.data.fill_(1)
 m.bias.data.zero_()

def KerasNet(input_shape=(224, 224, 3)):
image_input = keras.layers.Input(shape=input_shape)
# conv1
network = keras.layers.Conv2D(
32, (3, 3), strides=(2, 2), padding="valid")(image_input)
network = keras.layers.BatchNormalization(
trainable=False, fused=False)(network)
network = keras.layers.Activation("relu")(network)
network = keras.layers.MaxPool2D(pool_size=(2, 2), strides=(2, 2))(network)

# conv2
network = keras.layers.Conv2D(
64, (3, 3), strides=(1, 1), padding="valid")(network)
network = keras.layers.BatchNormalization(
trainable=False, fused=True)(network)
network = keras.layers.Activation("relu")(network)
network = keras.layers.MaxPool2D(pool_size=(2, 2), strides=(2, 2))(network)

model = keras.Model(inputs=image_input, outputs=network)

return model

class PytorchToKeras(object):
def __init__(self, pModel, kModel):
super(PytorchToKeras, self)
self.__source_layers = []
self.__target_layers = []
self.pModel = pModel
self.kModel = kModel
tf.keras.backend.set_learning_phase(0)

def __retrieve_k_layers(self):
for i, layer in enumerate(self.kModel.layers):
 if len(layer.weights) > 0:
 self.__target_layers.append(i)

def __retrieve_p_layers(self, input_size):

input = torch.randn(input_size)
input = Variable(input.unsqueeze(0))
hooks = []

def add_hooks(module):

def hook(module, input, output):
 if hasattr(module, "weight"):
  # print(module)
  self.__source_layers.append(module)

if not isinstance(module, nn.ModuleList) and not isinstance(module, nn.Sequential) and module != self.pModel:
 hooks.append(module.register_forward_hook(hook))

self.pModel.apply(add_hooks)

self.pModel(input)
for hook in hooks:
 hook.remove()

def convert(self, input_size):
self.__retrieve_k_layers()
self.__retrieve_p_layers(input_size)

for i, (source_layer, target_layer) in enumerate(zip(self.__source_layers, self.__target_layers)):
 print(source_layer)
 weight_size = len(source_layer.weight.data.size())
 transpose_dims = []
 for i in range(weight_size):
 transpose_dims.append(weight_size - i - 1)
 if isinstance(source_layer, nn.Conv2d):
 transpose_dims = [2,3,1,0]
 self.kModel.layers[target_layer].set_weights([source_layer.weight.data.numpy(
 ).transpose(transpose_dims), source_layer.bias.data.numpy()])
 elif isinstance(source_layer, nn.BatchNorm2d):
 self.kModel.layers[target_layer].set_weights([source_layer.weight.data.numpy(), source_layer.bias.data.numpy(),
       source_layer.running_mean.data.numpy(), source_layer.running_var.data.numpy()])
def save_model(self, output_file):
self.kModel.save(output_file)

def save_weights(self, output_file):
self.kModel.save_weights(output_file, save_format='h5')

pytorch_model = PytorchNet()
keras_model = KerasNet(input_shape=(224, 224, 3))

torch.save(pytorch_model, 'test.pth')

#Load the pretrained model
pytorch_model = torch.load('test.pth')

# #Time to transfer weights
converter = PytorchToKeras(pytorch_model, keras_model)
converter.convert((3, 224, 224))

# #Save the converted keras model for later use
# converter.save_weights("keras.h5")
converter.save_model("keras_model.h5")

# convert keras model to tflite model
converter = tf.contrib.lite.TocoConverter.from_keras_model_file(
"keras_model.h5")
tflite_model = converter.convert()
open("convert_model.tflite", "wb").write(tflite_model)

补充知识：tensorflow模型转换成tensorflow lite模型

1.把graph和网络模型打包在一个文件中

bazel build tensorflow/python/tools:freeze_graph && \
bazel-bin/tensorflow/python/tools/freeze_graph \
--input_graph=eval_graph_def.pb \
--input_checkpoint=checkpoint \
--output_graph=frozen_eval_graph.pb \
--output_node_names=outputs

For example:

bazel-bin/tensorflow/python/tools/freeze_graph \
--input_graph=./mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_eval.pbtxt \
--input_checkpoint=./mobilenet_v1_1.0_224/mobilenet_v1_1.0_224.ckpt \
--output_graph=./mobilenet_v1_1.0_224/frozen_eval_graph_test.pb \
--output_node_names=MobilenetV1/Predictions/Reshape_1

2.把第一步中生成的tensorflow pb模型转换为tf lite模型

转换前需要先编译转换工具

bazel build tensorflow/contrib/lite/toco:toco

转换分两种，一种的转换为float的tf lite，另一种可以转换为对模型进行unit8的量化版本的模型。两种方式如下：

非量化的转换：

./bazel-bin/third_party/tensorflow/contrib/lite/toco/toco \ 官网给的这个路径不对      
./bazel-bin/tensorflow/contrib/lite/toco/toco \        
—input_file=./mobilenet_v1_1.0_224/frozen_eval_graph_test.pb \  
—output_file=./mobilenet_v1_1.0_224/tflite_model_test.tflite \  
--input_format=TENSORFLOW_GRAPHDEF --output_format=TFLITE \      
--inference_type=FLOAT \          
--input_shape="1,224, 224,3" \          
--input_array=input \            
--output_array=MobilenetV1/Predictions/Reshape_1

量化方式的转换(注意，只有量化训练的模型才能进行量化的tf_lite转换)：

./bazel-bin/third_party/tensorflow/contrib/lite/toco/toco \
./bazel-bin/tensorflow/contrib/lite/toco/toco \
--input_file=frozen_eval_graph.pb \
--output_file=tflite_model.tflite \
--input_format=TENSORFLOW_GRAPHDEF --output_format=TFLITE \
--inference_type=QUANTIZED_UINT8 \
--input_shape="1,224, 224,3" \
--input_array=input \
--output_array=outputs \
--std_value=127.5 --mean_value=127.5

来源：https://blog.csdn.net/computerme/article/details/84144930