我有用于卷积神经网络的代码。
convnet = input_data(shape=[None, IMG_SIZE, IMG_SIZE, IMAGE_CHANNELS], name='input')
convnet1 = conv_2d(convnet, FIRST_NUM_CHANNEL, FILTER_SIZE, activation='relu')
convnet1 = max_pool_2d(convnet1, FILTER_SIZE)
convnet2 = conv_2d(convnet1, FIRST_NUM_CHANNEL*2, FILTER_SIZE, activation='relu')
convnet2 = max_pool_2d(convnet2, FILTER_SIZE)
convnet_final = fully_connected(convnet2, FIRST_NUM_CHANNEL*16, activation='relu')
convnet_final = dropout(convnet_final, 0.8)
convnet_final = fully_connected(convnet_final, NUM_OUTPUT, activation='softmax')
convnet_final = regression(convnet_final, optimizer='adam', learning_rate=LR, loss='categorical_crossentropy', name='targets')
model = tflearn.DNN(convnet_final, tensorboard_dir='log')
如何可视化convnet1或convnet2？我已经有了以下代码：
img = convnet1[0, :, :, 0]
但是 img 的数据类型为 Tensor（“ MaxPool2D / MaxPool：0”，shape =（?, 52，52，32），dtype = float32） 。我仍然需要将其转换为numpy数组之类的东西，或者可以在 imshow（）或任何其他图像显示中传递的东西。我还可以通过使用 model.session 获得运行该模型的会话。