Hello,
I'm trying to use PredNet on a dataset with binary images (black/white).
In the training file I changed the number of channels to 1 instead of 3.
However, an error shows up when I try to train it.
File "t1_train.py", line 78, in <module> validation_data=val_generator, validation_steps=N_seq_val / batch_size)...... raise TypeError(msg % (v1.type, v2.type, a1, b1, b2, a2)) TypeError: The broadcast pattern of the output of scan (TensorType(float32, 5D)) is inconsistent with the one provided inoutput_info(TensorType(float32, (False, False, False, True))). The output on axis 3 isFalse, but it isTrueon axis 4 inoutput_info. This can happen if one of the dimension is fixed to 1 in the input, while it is still variable in the output, or vice-verca. You have to make them consistent, e.g. using theano.tensor.{patternbroadcast,unbroadcast,addbroadcast}.
As far as I understand, one of the dimensions of the input and output doesn't match. I don't really find the way to modify the output dimensions of the network. How can I do to make it to modify the network and work with binary images?
Hello, I'm trying to use PredNet on a dataset with binary images (black/white). In the training file I changed the number of channels to 1 instead of 3. However, an error shows up when I try to train it.
File "t1_train.py", line 78, in <module> validation_data=val_generator, validation_steps=N_seq_val / batch_size)......raise TypeError(msg % (v1.type, v2.type, a1, b1, b2, a2)) TypeError: The broadcast pattern of the output of scan (TensorType(float32, 5D)) is inconsistent with the one provided inoutput_info(TensorType(float32, (False, False, False, True))). The output on axis 3 isFalse, but it isTrueon axis 4 inoutput_info. This can happen if one of the dimension is fixed to 1 in the input, while it is still variable in the output, or vice-verca. You have to make them consistent, e.g. using theano.tensor.{patternbroadcast,unbroadcast,addbroadcast}.As far as I understand, one of the dimensions of the input and output doesn't match. I don't really find the way to modify the output dimensions of the network. How can I do to make it to modify the network and work with binary images?