RuntimeError: Sizes of tensors must match except in dimension 1.

[cta5425]

Hi Ulrich,

Many thanks for your contribution! I have a question regarding the methodology in general and another more specific question. Why did you choose to combine CINN with Variational Autoencoder and did not use a conditional variational Autoencoder?

Also, I tried to run the jupyter notebooks provided, they all worked fine, except in pose estimation example in pose_estimation.ipynb. It occured on the line: zz = torch.randn(batch_size, 12).to(device). I saw in some other commits, that you sat the batch_size = 1, when I do the same I avoid the error but the predictions are far away from true. When the batch_size = 32 as it was, I get the following error:

`
Ground truth pose:
tensor([[-0.9466, 0.0161, -0.3221, -1.1683],
[-0.3225, -0.0472, 0.9454, 3.4293],
[ 0.0000, 0.9988, 0.0498, 0.1808],
[ 0.0000, 0.0000, 0.0000, 1.0000]], device='cuda:0')

RuntimeError Traceback (most recent call last)
Cell In [13], line 99
97 # ------ Finally, we compare the results ------
98 print("Ground truth pose:\n",new_pose)
---> 99 pred_pose = tau.reverse(zz, z).squeeze(-1).squeeze(-1)
100 pred_pose = to_pose(pred_pose)
101 print("---------------\n Predicted pose: \n", pred_pose)

File ~/local-repo/AutoNeRF/models/cinn.py:80, in ConditionalTransformer.reverse(self, out, conditioning)
78 def reverse(self, out, conditioning):
79 embedding = self.embed(conditioning)
---> 80 return self.flow(out, embedding, reverse=True)

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
1530 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1531 else:
-> 1532 return self._call_impl(*args, **kwargs)

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
1536 # If we don't have any hooks, we want to skip the rest of the logic in
1537 # this function, and just call forward.
1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1539 or _global_backward_pre_hooks or _global_backward_hooks
1540 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541 return forward_call(*args, **kwargs)
1543 try:
1544 result = None

File ~/local-repo/AutoNeRF/models/blocks.py:326, in ConditionalFlow.forward(self, x, embedding, reverse)
324 else:
325 for i in reversed(range(self.n_flows)):
--> 326 x = self.sub_layers[i](x, hconds[i], reverse=True)
327 return x

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
1530 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1531 else:
-> 1532 return self._call_impl(*args, **kwargs)

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
1536 # If we don't have any hooks, we want to skip the rest of the logic in
1537 # this function, and just call forward.
1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1539 or _global_backward_pre_hooks or _global_backward_hooks
1540 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541 return forward_call(*args, **kwargs)
1543 try:
1544 result = None

File ~/local-repo/AutoNeRF/models/blocks.py:235, in ConditionalFlatDoubleCouplingFlowBlock.forward(self, x, xcond, reverse)
233 h = x
234 h = self.shuffle(h, reverse=True)
--> 235 h = self.coupling(h, xcond, reverse=True)
236 h = self.activation(h, reverse=True)
237 h = self.norm_layer(h, reverse=True)

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
1530 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1531 else:
-> 1532 return self._call_impl(*args, **kwargs)

File ~/anaconda3/envs/nerfstudio/lib/python3.8/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
1536 # If we don't have any hooks, we want to skip the rest of the logic in
1537 # this function, and just call forward.
1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1539 or _global_backward_pre_hooks or _global_backward_hooks
1540 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541 return forward_call(*args, **kwargs)
1543 try:
1544 result = None

File ~/local-repo/AutoNeRF/models/blocks.py:110, in ConditionalDoubleVectorCouplingBlock.forward(self, x, xc, reverse)
108 x = torch.cat(torch.chunk(x, 2, dim=1)[::-1], dim=1)
109 x = torch.chunk(x, 2, dim=1)
--> 110 conditioner_input = torch.cat((x[idx_apply], xc), dim=1)
111 x_ = (x[idx_keep] - self.ti) * self.si.neg().exp()
112 x = torch.cat((x[idx_apply], x_), dim=1)

RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 32 but got size 1 for tensor number 1 in the list.
`

Many thanks in advance.