pytorch加载darknet权重文件

• 以yolov3-tiny为例，加载函数是ting.py的load_darknet
• base.py

import torch
import torch.nn as nn

class BaseConv(nn.Module):
	def __init__(self, in_channel, out_channel, k_size, stride, pad, bn, act):
		super().__init__()
		self.conv = nn.Conv2d(in_channel, out_channel, k_size, stride, pad, bias=not bn)
		if bn == True:
			self.bn = nn.BatchNorm2d(out_channel)
		else:
			self.bn = nn.Identity()
		if act == 'leaky': self.act = nn.LeakyReLU(0.1)
		elif act == 'linear': self.act = nn.Identity()
	def forward(self, x):
		x = self.conv(x)
		x = self.bn(x)
		return self.act(x)
		
if __name__ == '__main__':
	data=torch.zeros(2,3, 320,320)
	bc=BaseConv(3, 16, 3, 1, 1, False, 'leaky')
	x = bc(data)
	print(x.shape)

• tiny.py（网络结构）

import torch
import torch.nn as nn
from base import BaseConv
import numpy as np

class Backbone(nn.Module):
	def __init__(self):
		super().__init__()
		self.conv0 = BaseConv(3, 16, 3, 1, 1, True, 'leaky')
		self.maxpool1 = nn.MaxPool2d(2)
		self.conv2 = BaseConv(16, 32, 3, 1, 1, True, 'leaky')
		self.maxpool3 = nn.MaxPool2d(2)
		self.conv4 = BaseConv(32, 64, 3, 1, 1, True, 'leaky') 
		self.maxpool5 = nn.MaxPool2d(2)
		self.conv6 = BaseConv(64, 128, 3, 1, 1, True, 'leaky')
		self.maxpool7 = nn.MaxPool2d(2)
		self.conv8 = BaseConv(128, 256, 3, 1, 1, True, 'leaky')
		self.maxpool9 = nn.MaxPool2d(2)
		self.conv10 = BaseConv(256, 512, 3, 1, 1, True, 'leaky')
		self.pad = nn.ZeroPad2d((0,1,0,1))
		self.maxpool11 = nn.MaxPool2d(2, 1)
		self.conv12 = BaseConv(512, 1024, 3, 1, 1, True, 'leaky')
		self.conv13 = BaseConv(1024, 256, 1, 1, 0, True, 'leaky')
		self.conv14 = BaseConv(256, 512, 3, 1, 1, True, 'leaky')
	def forward(self, x):
		x0=self.conv0(x)
		x1=self.maxpool1(x0)  # 1    16*208*208
		x2=self.conv2(x1)
		x3=self.maxpool3(x2)  # 3    32*104*104
		x4=self.conv4(x3)
		x5=self.maxpool5(x4)  # 5    64*52*52
		x6=self.conv6(x5)
		x7=self