vgg11_transfer_to_vgg13

# 建構各層轉換式
def copy_weights(name, layer):
    if 'weight' in name:
        layer.weight.data.copy_(vgg11_state_dict[name + '.weight'])
    if 'bias' in name:
        layer.bias.data.copy_(vgg11_state_dict[name + '.bias'])
    # BN層除了weight和bias以外，還有running_mean、running_var、num_batches_tracked要複製
    if 'running_mean' in name:
        layer.running_mean.copy_(vgg11_state_dict[name + '.running_mean'])
    if 'running_var' in name:
        layer.running_var.copy_(vgg11_state_dict[name + '.running_var'])
    if 'num_batches_tracked' in name:
        layer.num_batches_tracked.copy_(vgg11_state_dict[name + '.num_batches_tracked'])

# VGG11和VGG13的結構(不含relu跟pooling)
# 默認基本結構會是conv→BN→conv→BN不斷循環，最後以BN作結
# 層數上，relu: +2  /  maxpooling: +1，所以像B, 128中間就是隔了relu+max所以+3
# 'VGG13': [ 64,  B,  64,  B, 128,  B, 128,  B, 256,  B, 256,  B, 512,  B, 512,  B, 512,  B, 512,  B]
# cnn_13 : [  0,  1,   3,  4,   7,  8,  10, 11,  14, 15,  17, 18,  21, 22,  24, 25,  28, 29,  31, 32]
# 'VGG11': [ 64,  B, 128,  B, 256,  B, 256,  B, 512,  B, 512,  B, 512,  B, 512,  B]
# cnn_11 : [  0,  1,   4,  5,   8,  9,  11, 12,  15, 16,  18, 19,  22, 23,  25, 26]
# 承上，13對11的對應關係為:
# vgg_13 0  1  7  8 14 15 17 18 21 22 24 25 28 29 31 32
# vgg_11 0  1  4  5  8  9 11 12 15 16 18 19 22 23 25 26

for name, layer in new_model.named_modules():
    # features的部分
    if 'features.0' in name:
        copy_weights('features.0',layer=layer)
    if 'features.1' in name:
        copy_weights('features.1',layer=layer)
    if 'features.7' in name:
        copy_weights('features.4',layer=layer)
    if 'features.8' in name:
        copy_weights('features.5',layer=layer)
    if 'features.14' in name:
        copy_weights('features.8',layer=layer)
    if 'features.15' in name:
        copy_weights('features.9',layer=layer)
    if 'features.17' in name:
        copy_weights('features.11',layer=layer)
    if 'features.18' in name:
        copy_weights('features.12',layer=layer)
    if 'features.21' in name:
        copy_weights('features.15',layer=layer)
    if 'features.22' in name:
        copy_weights('features.16',layer=layer)
    if 'features.24' in name:
        copy_weights('features.18',layer=layer)
    if 'features.25' in name:
        copy_weights('features.19',layer=layer)
    if 'features.28' in name:
        copy_weights('features.22',layer=layer)
    if 'features.29' in name:
        copy_weights('features.23',layer=layer)
    if 'features.31' in name:
        copy_weights('features.25',layer=layer)
    if 'features.32' in name:
        copy_weights('features.26',layer=layer)
    # classifier的部分
    if 'classifier.0' in name:
        copy_weights('classifier.0',layer=layer)
    if 'classifier.3' in name:
        copy_weights('classifier.3',layer=layer)
    if 'classifier.6' in name:
        copy_weights('classifier.6',layer=layer)