vacant-ztz
commented
2 months ago hello，have you solved the problem? and how?
Open javierpastorfernandez opened 8 months ago
vacant-ztz
commented
2 months ago hello，have you solved the problem? and how?
epochs: 0%| | 0/5 [06:03<?, ?it/s, loss=6.67, lr=0.00112] Traceback (most recent call last): File "train.py", line 209, in
main()
File "train.py", line 168, in main
train_model(
File "/home/javpasto/Documents/sfd/tools/train_utils/train_utils.py", line 114, in train_model
accumulated_iter = train_one_epoch(
File "/home/javpasto/Documents/sfd/tools/train_utils/train_utils.py", line 51, in train_one_epoch
loss, tb_dict, disp_dict = model_func(model, batch)
File "/home/javpasto/Documents/sfd/pcdet/models/init.py", line 28, in model_func
ret_dict, tb_dict, disp_dict = model(batch_dict)
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, kwargs)
File "/home/javpasto/Documents/sfd/pcdet/models/detectors/sfd.py", line 13, in forward
batch_dict = cur_module(batch_dict)
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, *kwargs)
File "/home/javpasto/Documents/sfd/pcdet/models/roi_heads/sfd_head.py", line 674, in forward
points_features_expand = self.cpconvs_layer(points_features, points_neighbor)[1:]
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(input, kwargs)
File "/home/javpasto/Documents/sfd/pcdet/models/roi_heads/sfd_head_utils.py", line 68, in forward
pointnet1_out_fea = self.pointnet1_fea(pointnet1_in_fea).view(N, -1)
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, *kwargs)
File "/home/javpasto/Documents/sfd/pcdet/models/roi_heads/sfd_head_utils.py", line 17, in forward
x = F.relu(self.bn1(self.conv1(x)))
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(input, kwargs)
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/modules/batchnorm.py", line 131, in forward
return F.batch_norm(
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/functional.py", line 2054, in batch_norm
_verify_batch_size(input.size())
File "/home/javpasto/anaconda3/envs/sfd/lib/python3.8/site-packages/torch/nn/functional.py", line 2037, in _verify_batch_size
raise ValueError('Expected more than 1 value per channel when training, got input size {}'.format(size))
ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 12, 1])**
Upon reviewing the sizes of the tensors involved: (points_features) shape: torch.Size([1, 9]) (points_neighbor) shape: torch.Size([1, 9]) (points_features) shape: torch.Size([1, 9]) (pointnet1_in_fea) shape: torch.Size([1, 1, 6]) one can see that after running the function roicrop3d_gpu, no inner ROI point is retrieve from the pseudo point cloud which throws an error when forwarding batch layer.
I do not how to solve this issue. I have been debugging and trying different things for weeks. Thank you in advance!