我正在尝试将 RGB 图像从模拟器传递到我的自定义神经网络。在RGB生成源(模拟器)处,RGB图像的维度是(3,144,256)。
这就是我构建神经网络的方式:
rgb_model = Sequential()
rgb = env.shape() // this is (3, 144, 256)
rgb_shape = (1,) + rgb
rgb_model.add(Conv2D(96, (11, 11), strides=(3, 3), padding='valid', activation='relu', input_shape=rgb_shape, data_format = "channels_first"))
Now, my rbg_shape is (1, 3, 144, 256).
这是我得到的错误:
rgb_model.add(Conv2D(96, (11, 11), strides=(3, 3), padding='valid', activation='relu', input_shape=rgb_kshape, data_format = "channels_first"))
File "/usr/local/lib/python2.7/dist-packages/keras/engine/sequential.py", line 166, in add
layer(x)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/base_layer.py", line 414, in call
self.assert_input_compatibility(inputs)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/base_layer.py", line 311, in assert_input_compatibility
str(K.ndim(x)))
ValueError: Input 0 is incompatible with layer conv2d_1: expected ndim=4, found ndim=5
为什么我的实际尺寸是 4 时 keras 会提示预期尺寸是 5?
P.S:我和this有同样的问题问题。理想情况下,我想对该帖子发表评论,但没有足够的声誉。
编辑:
这是处理错误的代码:
rgb_shape = env.rgb.shape
rgb_model = Sequential()
rgb_model.add(Conv2D(96, (11, 11), strides=(3, 3), padding='valid', activation='relu', input_shape=rgb_shape, data_format = "channels_first"))
rgb_model.add(Conv2D(128, (3, 3), strides=(2, 2), padding='valid', activation='relu', data_format = "channels_first" ))
rgb_model.add(Conv2D(384, (3, 3), strides=(1, 1), padding='valid', activation='relu', data_format = "channels_first"))
rgb_model.add(Conv2D(384, (3, 3), strides=(1,1), padding='valid', activation='relu', data_format = "channels_first"))
rgb_model.add(Conv2D(256, (3,3), strides=(1,1), padding='valid', activation='relu', data_format = "channels_first"))
rgb_model.add(Flatten())
rgb_input = Input(shape=rgb_shape)
rgb = rgb_model(rgb_input)
这是我在 Conv2D 中将 env.rgb.shape 作为 input_shape 传递时出现的新错误:
dqn.fit(env, callbacks=callbacks, nb_steps=250000, visualize=False, verbose=0, log_interval=100)
File "/usr/local/lib/python2.7/dist-packages/rl/core.py", line 169, in fit
action = self.forward(observation)
File "/usr/local/lib/python2.7/dist-packages/rl/agents/dqn.py", line 228, in forward
q_values = self.compute_q_values(state)
File "/usr/local/lib/python2.7/dist-packages/rl/agents/dqn.py", line 69, in compute_q_values
q_values = self.compute_batch_q_values([state]).flatten()
File "/usr/local/lib/python2.7/dist-packages/rl/agents/dqn.py", line 64, in compute_batch_q_values
q_values = self.model.predict_on_batch(batch)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1276, in predict_on_batch
x, _, _ = self._standardize_user_data(x)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 754, in _standardize_user_data
exception_prefix='input')
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training_utils.py", line 126, in standardize_input_data
'with shape ' + str(data_shape))
ValueError: Error when checking input: expected input_1 to have 4 dimensions, but got array with shape (1, 1, 3, 144, 256)
请您参考如下方法:
Conv2D 层的输入形状是(num_channels, width, height)。所以你不应该添加另一个维度(实际上输入的形状是 (batch_size, num_channels, width, height) 但是你不需要在这里设置 batch_size ;它将是在 fit 方法中设置)。只需将 input_shape=env.shape 传递给 Conv2D 即可正常工作。
编辑:
为什么要定义一个Input 层并将其传递给模型?它不是这样工作的。首先,您需要使用compile 方法编译模型,然后使用fit 方法在训练数据上进行训练,然后使用predict 方法使预测。我强烈建议阅读 official guide找出这些东西是如何工作的。
