根據Kevin 訓練VGG教程而來

假設我們這裡已經有了一個VGG16層的引數原始檔案，想要開啟來看一下引數設定，使用以下函式就行了，涉及到的庫自行補充

測試載入進來的引數
def test_load():
    data_path = './/VGG-pretrain//vgg16.npy'  # 檔案儲存路徑
    # 注意這個檔案要到網上自行下載
    data_dict = np.load(data_path, encoding='latin1').item()
    keys = sorted(data_dict.keys())
    for key in keys:
        weights = data_dict[key][0]
        biases = data_dict[key][1]
        print('\n')
        print(key)
        print('weights shape: ', weights.shape)
        print('biases shape: ', biases.shape)
 

結果展示：

conv1_1

weights shape:  (3, 3, 3, 64)
biases shape:  (64,)




conv1_2
weights shape:  (3, 3, 64, 64)
biases shape:  (64,)




conv2_1
weights shape:  (3, 3, 64, 128)
biases shape:  (128,)




conv2_2
weights shape:  (3, 3, 128, 128)
biases shape:  (128,)




conv3_1
weights shape:  (3, 3, 128, 256)
biases shape:  (256,)




conv3_2
weights shape:  (3, 3, 256, 256)
biases shape:  (256,)




conv3_3
weights shape:  (3, 3, 256, 256)
biases shape:  (256,)




conv4_1
weights shape:  (3, 3, 256, 512)
biases shape:  (512,)




conv4_2
weights shape:  (3, 3, 512, 512)
biases shape:  (512,)




conv4_3
weights shape:  (3, 3, 512, 512)
biases shape:  (512,)




conv5_1
weights shape:  (3, 3, 512, 512)
biases shape:  (512,)




conv5_2
weights shape:  (3, 3, 512, 512)
biases shape:  (512,)




conv5_3
weights shape:  (3, 3, 512, 512)
biases shape:  (512,)




fc6
weights shape:  (25088, 4096)
biases shape:  (4096,)




fc7
weights shape:  (4096, 4096)
biases shape:  (4096,)




fc8
weights shape:  (4096, 1000)
biases shape:  (1000,)