Problem with hidden layers

[lsteffenel]

Dear developers, I'm trying to use predrnn-pytorch but I face the following error each time I declare hidden layers mimicking an Unet architecture ("64,32,64", for example). No error happen if the hidden layers are regular ("64,64,64").

Do you have a hint on how to solve this ?

best regards.

Traceback (most recent call last):
File "run.py", line 223, in
train_wrapper(model)
File "run.py", line 191, in train_wrapper
trainer.train(model, ims, real_input_flag, args, itr)
File "/notebooks/predrnn-pytorch/core/trainer.py", line 15, in train
cost = model.train(ims, real_input_flag)
File "/notebooks/predrnn-pytorch/core/models/model_factory.py", line 42, in train
next_frames, loss = self.network(frames_tensor, mask_tensor)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/notebooks/predrnn-pytorch/core/models/predrnn_v2.py", line 92, in forward
h_t[i], c_t[i], memory, delta_c, delta_m = self.cell_list[i](h_t[i - 1], h_t[i], c_t[i], memory)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/notebooks/predrnn-pytorch/core/layers/SpatioTemporalLSTMCell_v2.py", line 49, in forward
m_concat = self.conv_m(m_t)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/container.py", line 204, in forward
input = module(input)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/conv.py", line 463, in forward
return self._conv_forward(input, self.weight, self.bias)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Given groups=1, weight of size [96, 32, 3, 3], expected input[8, 64, 32, 32] to have 32 channels, but got 64 channels instead

[wuhaixu2016]

Hi, sorry for this bug. The repo can only support the configuration that all the layers contain the same hidden channels. You need to change the layer module to ensure the U-Net style model.