MXNet 2 pytorch 模型转换

1、先定义好pytorch的网络结构： 没怎么接触人脸识别  insightface提供r50 里是 IResBlock，第一个卷积还是3x3 而非7x7

# -*- coding: utf-8 -*-
"""
Created on 18-5-21 下午5:26
@author: ronghuaiyang
"""
import torch
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
import torch.nn.utils.weight_norm as weight_norm
import torch.nn.functional as F


# __all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101',
#            'resnet152']


model_urls = {
    'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
    'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
    'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
    'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
    'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
}


def conv3x3(in_planes, out_planes, stride=1):
    """3x3 convolution with padding"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
                     padding=1, bias=False)


class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out


class IRBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None, use_se=True):
        super(IRBlock, self).__init__()
        self.bn1 = nn.BatchNorm2d(inplanes)
        self.conv1 = conv3x3(inplanes, planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.relu1 = nn.LeakyReLU(0.25)#nn.PReLU()
        self.conv2 = conv3x3(planes, planes, stride)
        self.bn3 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride
        self.use_se = use_se
        if self.use_se:
            self.se = SEBlock(planes)

    def forward(self, x):
        residual = x
        out = self.bn0(x)
        out = self.conv1(out)
        out = self.bn1(out)
        out = self.prelu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        if self.use_se:
            out = self.se(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.prelu(out)

        return out


class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
                               padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out


class SEBlock(nn.Module):
    def __init__(self, channel, reduction=16):
        super(SEBlock, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
                nn.Linear(channel, channel // reduction),
                nn.PReLU(),
                nn.Linear(channel // reduction, channel),
                nn.Sigmoid()
        )

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y


class ResNetFace(nn.Module):
    def __init__(self, block, layers, use_se=False):
        self.inplanes = 64
        self.use_se = use_se
        super(ResNetFace, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, padding=1,stride=2, bias=False)
        self.bn0 = nn.BatchNorm2d(64)
        self.prelu = nn.LeakyReLU(0.25)#nn.PReLU()
        self.maxpool = nn.MaxPool2d(kernel_size=2, stride=2)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.bn1 = nn.BatchNorm2d(512)
        self.dropout = nn.Dropout()
        self.pre_fc1 = nn.Linear(512 * 7 * 7, 512)
        self.fc1 = nn.BatchNorm1d(512)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.xavier_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d) or isinstance(m, nn.BatchNorm1d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.xavier_normal_(m.weight)
                nn.init.constant_(m.bias, 0)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )
        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample, use_se=self.use_se))
        self.inplanes = planes
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes, use_se=self.use_se))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.prelu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.bn4(x)
        x = self.dropout(x)
        x = x.view(x.size(0), -1)
        x = self.fc5(x)
        x = self.bn5(x)

        return x


class ResNet(nn.Module):

    def __init__(self, block, layers):
        self.inplanes = 64
        super(ResNet, self).__init__()
        # self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
        #                        bias=False)
        self.conv1 = nn.Conv2d(1, 64, kernel_size=3, stride=1, padding=1,
                               bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        # self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0], stride=2)
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        # self.avgpool = nn.AvgPool2d(8, stride=1)
        # self.fc = nn.Linear(512 * block.expansion, num_classes)
        self.fc5 = nn.Linear(512 * 8 * 8, 512)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        # x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        # x = nn.AvgPool2d(kernel_size=x.size()[2:])(x)
        # x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.fc5(x)

        return x


def resnet18(pretrained=False, **kwargs):
    """Constructs a ResNet-18 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
    return model


def resnet34(pretrained=False, **kwargs):
    """Constructs a ResNet-34 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet34']))
    return model


def resnet50(pretrained=False, **kwargs):
    """Constructs a ResNet-50 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
    return model


def resnet101(pretrained=False, **kwargs):
    """Constructs a ResNet-101 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet101']))
    return model


def resnet152(pretrained=False, **kwargs):
    """Constructs a ResNet-152 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
    if pretrained:
        model.load_state_dict(model_zoo.load_url(model_urls['resnet152']))
    return model


def resnet_face18(use_se=True, **kwargs):
    model = ResNetFace(IRBlock, [2, 2, 2, 2], use_se=use_se, **kwargs)
    return model

　　

2、转换代码

mxnet模型里json文件记录了模型相关的键名，

import torch.nn as nn
import torch.utils.checkpoint as cp
import mxnet as mx 
import torch
from  resnet import ResNetFace,IRBlock
archs = ['iresnet50']
mx_model = './model-r50-am-lfw/model'
mx_epoch = 14

_,arg_params, aux_params = mx.model.load_checkpoint(mx_model,mx_epoch)

class convert_model(object):

    def init_model(self, model,param_dict,aux_params):
        # print(model)
        layer=[3,4,14,3]
        for n, m in model.named_modules():
            print(n)
            n1 = n.split('.')
            if len(n1)>2:
                stage = n1[0][-1]
                unit = int(n1[1])+1
                op = n1[2]
                print(stage,unit,op)
                if op=='relu':
                     op = op+'1'
                if op == 'downsample':
                     r='0000'
                     if len(n1)>3:
                         r =n1[3]
                     if r=='0':
                         op = 'conv1sc'
                     else:
                         op = 'sc'
    #
#            if op == ''
                n = 'stage'+stage+'_unit'+str(unit)+'_'+op
                print(n)
            if n=='conv1':
                self.conv1_init(n,m,param_dict)
            elif isinstance(m, nn.BatchNorm2d):
                if n=='bn0':
                    self.bn_init('bn0', m, param_dict,aux_params)
                else:
                    self.bn_init(n, m, param_dict,aux_params)
            elif isinstance(m, nn.Conv2d):

                self.conv_init(n, m, param_dict)
            elif isinstance(m, nn.Linear):
                self.fc_init(n, m, param_dict)
            elif isinstance(m, nn.PReLU):
                if n=='prelu':
                    self.prelu_init('relu0', m, param_dict)
                else:

                    self.prelu_init(n, m, param_dict)
 
 
        return model
 
    def bn_init(self, n, m, param_dict,aux_params):
        print(torch.FloatTensor(param_dict[n+'_gamma'].asnumpy()).size())
        if not (m.weight is None):
            m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_gamma'].asnumpy()))
            m.bias.data.copy_(torch.FloatTensor(param_dict[n+'_beta'].asnumpy()))
        m.running_mean.copy_(torch.FloatTensor(aux_params[n+'_moving_mean'].asnumpy()))
        m.running_var.copy_(torch.FloatTensor(aux_params[n+'_moving_var'].asnumpy()))
    def conv1_init(self, n, m, param_dict):
     # print('n = ', n)
        #$n = 'conv0'
        a = torch.FloatTensor(param_dict['conv0'+'_weight'].asnumpy())
        print(a.size())
        print(m.weight.size())
        m.weight.data.copy_(torch.FloatTensor(param_dict['conv0'+'_weight'].asnumpy()))
        

    def conv_init(self, n, m, param_dict):
     # print('n = ', n)
        a = torch.FloatTensor(param_dict[n+'_weight'].asnumpy())
        print(a.size(),n)
        print(m.weight.size())
        m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_weight'].asnumpy()))
        #if n in ['conv1_1', 'conv4_1', 'conv3_1', 'conv2_1']:
        #m.bias.data.copy_(torch.FloatTensor(param_dict[n + '_bias'].asnumpy()))

    def fc_init(self, n, m, param_dict):
        m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_weight'].asnumpy()))
        m.bias.data.copy_(torch.FloatTensor(param_dict[n+'_bias'].asnumpy()))

    def prelu_init(self, n, m, param_dict):
        print(torch.FloatTensor(param_dict[n + '_gamma'].asnumpy()).size())
        m.weight.data.copy_(torch.FloatTensor(param_dict[n + '_gamma'].asnumpy()))

c = convert_model()
r50  = ResNetFace(IRBlock, [3,4,14,3])#getattr(models, arch)()#ResNet(BasicBlock,[3,4,14,3])
model = c.init_model(r50,arg_params,aux_params)
#r50.load_state_dict(torch.load('./r50.pkl'))
#exit(0)
torch.save(model.state_dict(),'./r50.pkl')

import torch.nn as nn
import torch.utils.checkpoint as cp
import mxnet as mx 
import torch
from  resnet import ResNetFace,IRBlock
archs = ['iresnet50']
mx_model = './model-r50-am-lfw/model'
mx_epoch = 14

_,arg_params, aux_params = mx.model.load_checkpoint(mx_model,mx_epoch)

class convert_model(object):

    def init_model(self, model,param_dict,aux_params):
        # print(model)
        layer=[3,4,14,3]
        for n, m in model.named_modules():
            print(n)
            n1 = n.split('.')
            if len(n1)>2:
                stage = n1[0][-1]
                unit = int(n1[1])+1
                op = n1[2]
                print(stage,unit,op)
                if op=='relu':
                     op = op+'1'
                if op == 'downsample':
                     r='0000'
                     if len(n1)>3:
                         r =n1[3]
                     if r=='0':
                         op = 'conv1sc'
                     else:
                         op = 'sc'
    #
#            if op == ''
                n = 'stage'+stage+'_unit'+str(unit)+'_'+op
                print(n)
            if n=='conv1':
                self.conv1_init(n,m,param_dict)
            elif isinstance(m, nn.BatchNorm2d):
                if n=='bn0':
                    self.bn_init('bn0', m, param_dict,aux_params)
                else:
                    self.bn_init(n, m, param_dict,aux_params)
            elif isinstance(m, nn.Conv2d):

                self.conv_init(n, m, param_dict)
            elif isinstance(m, nn.Linear):
                self.fc_init(n, m, param_dict)
            elif isinstance(m, nn.PReLU):
                if n=='prelu':
                    self.prelu_init('relu0', m, param_dict)
                else:

                    self.prelu_init(n, m, param_dict)
 
 
        return model
 
    def bn_init(self, n, m, param_dict,aux_params):
        print(torch.FloatTensor(param_dict[n+'_gamma'].asnumpy()).size())
        if not (m.weight is None):
            m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_gamma'].asnumpy()))
            m.bias.data.copy_(torch.FloatTensor(param_dict[n+'_beta'].asnumpy()))
        m.running_mean.copy_(torch.FloatTensor(aux_params[n+'_moving_mean'].asnumpy()))
        m.running_var.copy_(torch.FloatTensor(aux_params[n+'_moving_var'].asnumpy()))
    def conv1_init(self, n, m, param_dict):
     # print('n = ', n)
        #$n = 'conv0'
        a = torch.FloatTensor(param_dict['conv0'+'_weight'].asnumpy())
        print(a.size())
        print(m.weight.size())
        m.weight.data.copy_(torch.FloatTensor(param_dict['conv0'+'_weight'].asnumpy()))
        

    def conv_init(self, n, m, param_dict):
     # print('n = ', n)
        a = torch.FloatTensor(param_dict[n+'_weight'].asnumpy())
        print(a.size(),n)
        print(m.weight.size())
        m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_weight'].asnumpy()))
        #if n in ['conv1_1', 'conv4_1', 'conv3_1', 'conv2_1']:
        #m.bias.data.copy_(torch.FloatTensor(param_dict[n + '_bias'].asnumpy()))

    def fc_init(self, n, m, param_dict):
        m.weight.data.copy_(torch.FloatTensor(param_dict[n+'_weight'].asnumpy()))
        m.bias.data.copy_(torch.FloatTensor(param_dict[n+'_bias'].asnumpy()))

    def prelu_init(self, n, m, param_dict):
        print(torch.FloatTensor(param_dict[n + '_gamma'].asnumpy()).size())
        m.weight.data.copy_(torch.FloatTensor(param_dict[n + '_gamma'].asnumpy()))

c = convert_model()
r50  = ResNetFace(IRBlock, [3,4,14,3])#getattr(models, arch)()#ResNet(BasicBlock,[3,4,14,3])
model = c.init_model(r50,arg_params,aux_params)
#r50.load_state_dict(torch.load('./r50.pkl'))
#exit(0)
torch.save(model.state_dict(),'./r50.pkl')