【Pytorch】| weight文件转为.pt文件 个人数据集类别数量，不同模型的转换yolo 成功案例

引言

weight文件前4-5个int32字节是head信息。单独保存，后面是float32数据，对应bn（gamma beta runing_mean running_var）和conv的权重参数
按照cfg文件定义的网络结构 ，解析cfg文件后，得到module_def, module，索引权重位置。

前提网络熟悉

yolov3输出的网络module_def结构，一共107个

[{'type': 'convolutional', 'batch_normalize': 1, 'filters': 32, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 64, 'size': 3, 'stride': 2, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 32, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 64, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 3, 'stride': 2, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 64, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 64, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 2, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 2, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 1024, 'size': 3, 'stride': 2, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 1024, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 1024, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 1024, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 1024, 'size': 3, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'shortcut', 'from': [-3], 'activation': 'linear'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 1024, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 1024, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'filters': 512, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
 {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 1024, 'activation': 'leaky'}, 
 {'type': 'convolutional', 'batch_normalize': 0, 'size': 1, 'stride': 1, 'pad': 1, 'filters': 24, 'activation': 'linear'},
 {'type': 'yolo', 'mask': [6, 7, 8], 'anchors': array([[ 10.,  13.],
       [ 16.,  30.],
       [ 33.,  23.],
       [ 30.,  61.],
       [ 62.,  45.],
       [ 59., 119.],
       [116.,  90.],
       [156., 198.],
       [373., 326.]]), 'classes': 3, 'num': 9, 'jitter': '.3', 'ignore_thresh': '.5', 'truth_thresh': 1, 'random': 1}, 
       {'type': 'route', 'layers': [-4]}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
       {'type': 'upsample', 'stride': 2}, 
       {'type': 'route', 'layers': [-1, 61]}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 512, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 512, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 256, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 512, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 0, 'size': 1, 'stride': 1, 'pad': 1, 'filters': 24, 'activation': 'linear'}, 
       {'type': 'yolo', 'mask': [3, 4, 5], 'anchors': array([[ 10.,  13.],
       [ 16.,  30.],
       [ 33.,  23.],
       [ 30.,  61.],
       [ 62.,  45.],
       [ 59., 119.],
       [116.,  90.],
       [156., 198.],
       [373., 326.]]), 'classes': 3, 'num': 9, 'jitter': '.3', 'ignore_thresh': '.5', 'truth_thresh': 1, 'random': 1},
       {'type': 'route', 'layers': [-4]},
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
       {'type': 'upsample', 'stride': 2}, 
       {'type': 'route', 'layers': [-1, 36]},
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 256, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'},
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 256, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'filters': 128, 'size': 1, 'stride': 1, 'pad': 1, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 1, 'size': 3, 'stride': 1, 'pad': 1, 'filters': 256, 'activation': 'leaky'}, 
       {'type': 'convolutional', 'batch_normalize': 0, 'size': 1, 'stride': 1, 'pad': 1, 'filters': 24, 'activation': 'linear'}, 
       {'type': 'yolo', 'mask': [0, 1, 2], 'anchors': array([[ 10.,  13.],
       [ 16.,  30.],
       [ 33.,  23.],
       [ 30.,  61.],
       [ 62.,  45.],
       [ 59., 119.],
       [116.,  90.],
       [156., 198.],
       [373., 326.]]), 'classes': 3, 'num': 9, 'jitter': '.3', 'ignore_thresh': '.5', 'truth_thresh': 1, 'random': 1}]

对应的model结构 0-106

Darknet(
  (module_list): ModuleList(
    (0): Sequential(
      (Conv2d): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (1): Sequential(
      (Conv2d): Conv2d(32, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (2): Sequential(
      (Conv2d): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (3): Sequential(
      (Conv2d): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (4): WeightedFeatureFusion()
    (5): Sequential(
      (Conv2d): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (6): Sequential(
      (Conv2d): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (7): Sequential(
      (Conv2d): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (8): WeightedFeatureFusion()
    (9): Sequential(
      (Conv2d): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (10): Sequential(
      (Conv2d): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (11): WeightedFeatureFusion()
    (12): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (13): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (14): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (15): WeightedFeatureFusion()
    (16): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (17): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (18): WeightedFeatureFusion()
    (19): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (20): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (21): WeightedFeatureFusion()
    (22): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (23): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (24): WeightedFeatureFusion()
    (25): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (26): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (27): WeightedFeatureFusion()
    (28): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (29): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (30): WeightedFeatureFusion()
    (31): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (32): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (33): WeightedFeatureFusion()
    (34): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (35): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (36): WeightedFeatureFusion()
    (37): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (38): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (39): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (40): WeightedFeatureFusion()
    (41): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (42): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (43): WeightedFeatureFusion()
    (44): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (45): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (46): WeightedFeatureFusion()
    (47): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (48): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (49): WeightedFeatureFusion()
    (50): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (51): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (52): WeightedFeatureFusion()
    (53): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (54): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (55): WeightedFeatureFusion()
    (56): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (57): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (58): WeightedFeatureFusion()
    (59): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (60): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (61): WeightedFeatureFusion()
    (62): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (63): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (64): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (65): WeightedFeatureFusion()
    (66): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (67): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (68): WeightedFeatureFusion()
    (69): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (70): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (71): WeightedFeatureFusion()
    (72): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (73): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (74): WeightedFeatureFusion()
    (75): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (76): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (77): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (78): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (79): Sequential(
      (Conv2d): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (80): Sequential(
      (Conv2d): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (81): Sequential(
      (Conv2d): Conv2d(1024, 24, kernel_size=(1, 1), stride=(1, 1))
    )
    (82): YOLOLayer()
    (83): FeatureConcat()
    (84): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (85): Upsample(scale_factor=2.0, mode=nearest)
    (86): FeatureConcat()
    (87): Sequential(
      (Conv2d): Conv2d(768, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (88): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (89): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (90): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (91): Sequential(
      (Conv2d): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (92): Sequential(
      (Conv2d): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (93): Sequential(
      (Conv2d): Conv2d(512, 24, kernel_size=(1, 1), stride=(1, 1))
    )
    (94): YOLOLayer()
    (95): FeatureConcat()
    (96): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (97): Upsample(scale_factor=2.0, mode=nearest)
    (98): FeatureConcat()
    (99): Sequential(
      (Conv2d): Conv2d(384, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (100): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (101): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (102): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (103): Sequential(
      (Conv2d): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (104): Sequential(
      (Conv2d): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (BatchNorm2d): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (activation): LeakyReLU(negative_slope=0.1)
    )
    (105): Sequential(
      (Conv2d): Conv2d(256, 24, kernel_size=(1, 1), stride=(1, 1))
    )
    (106): YOLOLayer()
  )
)

转换代码

项目下载

在u版的yolo目录下运行（包含所需的model）

weights2pt转换代码

以yolov3 yolov3-spp为例

import torch
#from models import Darknet, load_darknet_weights
from models import Darknet
import numpy as np
import os

def load_darknet_weights(self, weights, cutoff=-1):
    # Parses and loads the weights stored in 'weights'
    # cutoff: save layers between 0 and cutoff (if cutoff = -1 all are saved)
    weights_file = weights.split(os.sep)[-1]

    # Try to download weights if not available locally
#    if not os.path.isfile(weights):
#        try:
#            os.system('wget https://pjreddie.com/media/files/' + weights_file + ' -O ' + weights)
#        except IOError:
#            print(weights + ' not found.\nTry https://drive.google.com/drive/folders/1uxgUBemJVw9wZsdpboYbzUN4bcRhsuAI')

    # Establish cutoffs
    if weights_file == 'darknet53.conv.74':
        cutoff = 75
    elif weights_file == 'yolov3-tiny.conv.15':
        cutoff = 15

    # Open the weights file
    with open(weights, 'rb') as f:
        header = np.fromfile(f, dtype=np.int32, count=4)  #yolov3是4 yolov3-spp以上是5 First five are header values

        # Needed to write header when saving weights
        self.header_info = header

        self.seen = header[3]  # number of images seen during training
        weights = np.fromfile(f, dtype=np.float32)  # The rest are weights

    ptr = 0
    for i, (module_def, module) in enumerate(zip(self.module_defs[:cutoff], self.module_list[:cutoff])):
        if module_def['type'] == 'convolutional':
            conv_layer = module[0]
            if module_def['batch_normalize']:
                # Load BN bias, weights, running mean and running variance
                bn_layer = module[1]
                num_b = bn_layer.bias.numel()  # Number of biases
                # Bias
                bn_b = torch.from_numpy(weights[ptr:ptr + num_b]).view_as(bn_layer.bias)
                bn_layer.bias.data.copy_(bn_b)
                ptr += num_b
                # Weight
                bn_w = torch.from_numpy(weights[ptr:ptr + num_b]).view_as(bn_layer.weight)
                bn_layer.weight.data.copy_(bn_w)
                ptr += num_b
                # Running Mean
                bn_rm = torch.from_numpy(weights[ptr:ptr + num_b]).view_as(bn_layer.running_mean)
                bn_layer.running_mean.data.copy_(bn_rm)
                ptr += num_b
                # Running Var
                bn_rv = torch.from_numpy(weights[ptr:ptr + num_b]).view_as(bn_layer.running_var)
                bn_layer.running_var.data.copy_(bn_rv)
                ptr += num_b
            else:
                # Load conv. bias
                num_b = conv_layer.bias.numel()
                conv_b = torch.from_numpy(weights[ptr:ptr + num_b]).view_as(conv_layer.bias)
                conv_layer.bias.data.copy_(conv_b)
                ptr += num_b
            # Load conv. weights
            num_w = conv_layer.weight.numel()
            conv_w = torch.from_numpy(weights[ptr:ptr + num_w]).view_as(conv_layer.weight)
            conv_layer.weight.data.copy_(conv_w)
            ptr += num_w

    return cutoff

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

cfg = 'my_cfg/yolo-obj_3cls.cfg'
weights = 'weights/yolo-obj_3cls_18000.weights'

model = Darknet(cfg).to(device)
load_darknet_weights(model, weights)
chkpt = {'epoch': -1, 'best_loss': None, 'model': model.state_dict(), 'optimizer': None}
torch.save(chkpt, 'weights/yolo-obj_3cls.pt')

pt2weights

from models import * 注意：新建的pt2weights.py与models.py在同一目录下


cfg = 'my_cfg/yolo-obj_3cls.cfg'

model=Darknet(cfg) #训练时候用的cfg 改成自己.cfg的地址 


checkpoint = torch.load("weights/yolo-obj_3cls.pt",map_location='cpu')#导入训练好的.pt文件 改成自己的.pt文件地址
model.load_state_dict(checkpoint['model'])
save_weights(model,path='weights/bt2weights_obj_3cls.weights',cutoff=-1)#生成.weights文件 改成自己的地址

checkpint内容

{'epoch': -1, 'best_loss': None, 'model': OrderedDict([
('module_list.0.conv_0.weight', tensor([[[[ 1.22904e-01,  1.03453e-01,  8.72016e-02],
          [ 1.00648e-01,  7.50468e-02,  9.15262e-02],
          [-6.72630e-03, -2.11742e-02,  2.15247e-02]], [[-3.27195e-02,  1.09584e-01,  1.08887e-01]
          ,.....[-2.40979e-02,  6.60297e-02,  5.63852e-02],[-4.11258e-02, -1.59853e-02, -2.48185e-02]]]])), 
 ('module_list.0.batch_norm_0.weight', tensor([ 2.81073,  0.97984,  1.30070,  0.61605, -1.12916,  2.66054,  0.73914,  0.63000,  0.75663, -0.47399,  0.97683,  0.94648,  1.12235,  1.35054,  2.16467,  1.36777,  0.93587,  1.80257,  1.42253,  1.00391,  1.39467, -1.76530, -0.76111, -0.79642,  0.63124,  1.64442,  1.34087,  0.86794,  2.53687,  2.07082,  0.58116,  1.69326])), 
('module_list.0.batch_norm_0.bias', tensor([-4.21948, -1.02276, -2.11478,  1.32232, -0.79700, -3.23063, -0.32061,  0.02714,  1.69935,  0.75224, -0.87066,  3.32068,  2.53358, -2.02973, -3.66029, -1.94644,  3.56482, -1.57652, -1.98882, -0.84602, -2.13664, -2.77446,  1.45947,  0.27386,  1.32830, -2.41020,  2.64488, -1.26660, -2.85382, -2.76217, -0.07457, -2.27203])), 
('module_list.0.batch_norm_0.running_mean', tensor([ 3.47822e-01, -1.38476e-01,  1.28535e-01,  1.99359e-01, -1.43450e-01,  1.93027e-01, -1.26285e-01,  5.53209e-02, -2.10289e-01, -2.40412e-01, -8.84716e-02, -2.61738e-05, -2.29773e-02,  9.82087e-02,  2.56201e-01,  1.20729e-01, -1.49580e-01, -1.24053e-01,  1.10301e-01, -6.28857e-02, -1.15364e-01,  1.67073e-01, -2.27964e-01, -4.17056e-01, -4.56573e-02,  1.28036e-01,  1.88810e-02, -1.37764e-02, -2.29639e-01, -3.33596e-01, -1.09311e-01,  2.01629e-01])),
('module_list.0.batch_norm_0.running_var', tensor([0.07550, 0.01245, 0.01136, 0.02681, 0.01321, 0.02595, 0.01172, 0.02844, 0.03290, 0.03974, 0.00646, 0.00892, 0.01218, 0.00612, 0.04389, 0.00941, 0.02433, 0.01332, 0.00934, 0.00453, 0.00895, 0.01764, 0.03760, 0.11741, 0.03504, 0.01099, 0.01229, 0.00182, 0.03769, 0.07087, 0.01180, 0.02624])), ('module_list.0.batch_norm_0.num_batches_tracked', tensor(0)),
 ......