7. RuntimeError: Given groups=1, weight of size [8, 1, 1, 1], expected input[12, 6, 256, 704] to have 1 channels, but got 6 channels instead

[MaxLeo-LMZ]

This problem occurs when I run test.py:

outputs = single_gpu_test(model, data_loader) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/memory_profiler.py", line 1188, in wrapper val = prof(func)(*args, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/memory_profiler.py", line 761, in f return func(*args, **kwds) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/apis/test.py", line 13, in single_gpu_test result = model(return_loss=False, rescale=True, **data) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/mmcv/parallel/data_parallel.py", line 50, in forward return super().forward(*inputs, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 166, in forward return self.module(*inputs[0], **kwargs[0]) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 128, in new_func output = old_func(*new_args, **new_kwargs) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 253, in forward outputs = self.forward_single( File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 128, in new_func output = old_func(*new_args, **new_kwargs) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 301, in forward_single feature = self.extract_camera_features( File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 133, in extract_camera_features x = self.encoders["camera"]["vtransform"]( File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/vtransforms/depth_lss.py", line 100, in forward x = super().forward(*args, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 214, in new_func output = old_func(*new_args, **new_kwargs) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/vtransforms/base.py", line 349, in forward x = self.get_cam_feats(img, depth, mats_dict) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 214, in new_func output = old_func(*new_args, **new_kwargs) File "/home/mingzhong/PycharmProjects/bevfusion/bevfusion/mmdet3d/models/vtransforms/depth_lss.py", line 88, in get_cam_feats d = self.dtransform(d) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/container.py", line 141, in forward input = module(input) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 446, in forward return self._conv_forward(input, self.weight, self.bias) File "/home/mingzhong/anaconda3/envs/bevfusion/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 442, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: Given groups=1, weight of size [8, 1, 1, 1], expected input[6, 6, 256, 704] to have 1 channels, but got 6 channels instead

I'm using the latest code and using the nuscense dataset.

[Linkon87]

same problem @zhijian-liu

[ChunyuFeng]

same problem @zhijian-liu

[Ricardolyg]

i have the same problem

[AlexIlis]

I have the same problem here - using latest code on nuscenes dataset.

[quantumdotsss]

Same here

[AlexIlis]

mmdet3d/models/vtransforms/base: line 38: add_depth_features=True -> False works for me. Thanks @JunShao0104

[Linkon87]

mmdet3d/models/vtransforms/base: line 38: add_depth_features=True -> False works for me. Thanks @JunShao0104

Thank you. However, this will make val results worse , wonder is there any other solution?

Try to use old branch may solve this , not the latest code

[Ricardolyg]

mmdet3d/models/vtransforms/base: line 38: add_depth_features=True -> False works for me. Thanks @JunShao0104

This works for me, but I get a new error when I run test. I feel this error has something to do with "add_depth_features=True -> False"

[Ricardolyg]

mmdet3d/models/vtransforms/base: line 38: add_depth_features=True -> False works for me. Thanks @JunShao0104

This works for me, but I get a new error when I run test. I feel this error has something to do with "add_depth_features=True -> False"

Traceback (most recent call last): File "tools/test.py", line 230, in main() File "tools/test.py", line 203, in main outputs = multi_gpu_test(model, data_loader, args.tmpdir, args.gpu_collect) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/mmdet/apis/test.py", line 98, in multi_gpu_test result = model(return_loss=False, rescale=True, **data) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 886, in forward output = self.module(*inputs[0], **kwargs[0]) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 128, in new_func output = old_func(*new_args, **new_kwargs) File "/home/dell/ricardo/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 253, in forward outputs = self.forward_single( File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 128, in new_func output = old_func(*new_args, **new_kwargs) File "/home/dell/ricardo/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 301, in forward_single feature = self.extract_camera_features( File "/home/dell/ricardo/bevfusion/mmdet3d/models/fusion_models/bevfusion.py", line 133, in extract_camera_features x = self.encoders["camera"]["vtransform"]( File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/home/dell/ricardo/bevfusion/mmdet3d/models/vtransforms/depth_lss.py", line 100, in forward x = super().forward(*args, **kwargs) File "/home/dell/anaconda3_ricardo/envs/bevfusion06/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 214, in new_func output = old_func(*new_args, **new_kwargs) File "/home/dell/ricardo/bevfusion/mmdet3d/models/vtransforms/base.py", line 291, in forward cur_coords = torch.inverse(cur_lidar_aug_matrix[:3, :3]).matmul( RuntimeError: cusolver error: CUSOLVER_STATUS_INTERNAL_ERROR, when calling cusolverDnCreate(handle)

Primary job terminated normally, but 1 process returned a non-zero exit code. Per user-direction, the job has been aborted.

---

mpirun detected that one or more processes exited with non-zero status, thus causing the job to be terminated. The first process to do so was:

Process name: [[60520,1],0] Exit code: 1

[Ricardolyg]

Amazing, I took the mmdet3d/models/vtransforms/base: line 38: "add_depth_features=False" changed back to true, and then I ran test as before, and then I changed back to False, and both errors disappeared, and test ran successfully.

[JunShao0104]

Hey guys, I think I have solved the problem of #455 and #478. For both evaluation and training, we should set not only the add_depth_features=False but height_expand=False at line 37 and line 38 in bevfusion/mmdet3d/models/vtransforms/base.py. These two flags, from my point of view, are for the Camera+Radar config. If these two flags are set to False, there will be no difference between the latest code and the stable old branch. In this way, the evaluation of the provided checkpoint (L+C) will be the same as the reported values.

[YueWangTO]

Hey guys, I think I have solved the problem of #455 and #478. For both evaluation and training, we should set not only the add_depth_features=False but height_expand=False at line 37 and line 38 in bevfusion/mmdet3d/models/vtransforms/base.py. These two flags, from my point of view, are for the Camera+Radar config. If these two flags are set to False, there will be no difference between the latest code and the stable old branch. In this way, the evaluation of the provided checkpoint (L+C) will be the same as the reported values.

I did as you said, set the two parameters of line 37 and 38 to false, but the training result of c+ l is still worse than the official result, may I ask you if you used the original pre-training model?

[JunShao0104]

Hey guys, I think I have solved the problem of #455 and #478. For both evaluation and training, we should set not only the add_depth_features=False but height_expand=False at line 37 and line 38 in bevfusion/mmdet3d/models/vtransforms/base.py. These two flags, from my point of view, are for the Camera+Radar config. If these two flags are set to False, there will be no difference between the latest code and the stable old branch. In this way, the evaluation of the provided checkpoint (L+C) will be the same as the reported values.

I did as you said, set the two parameters of line 37 and 38 to false, but the training result of c+ l is still worse than the official result, may I ask you if you used the original pre-training model?

Check the number of LiDAR sweeps in the training procedure. It should be 9 instead of 0. In this way, the re-training of the official BEVFusion LC branch will be correct.

[YueWangTO]

感谢您的来信，已收到

[wstzx]

Hello author, thank you so much for open sourcing your code. I tried to run your latest code using nuscenes-mini on a single GPU on 3090 and A5000, and it showed: 7.RuntimeError: Given groups=1, weight of size [8, 1, 1, 1], expected input[12, 6, 256, 704] to have 1 channels, but got 6 channels instead , so I setmmdet3d/models/vtransforms/base line37 and line38: False, but the results of the test MAP = 58, and after I used the two other old version to test, the result is only 58, far lower than the final result, is the reason I used one GPU?Here are the results of my single GPU tests in nuscenes-mini:

[YueWangTO]

感谢您的来信，已收到

[wstzx]

Hello author, thank you so much for open sourcing your code. I tried to run your latest code using nuscenes-mini on a single GPU on 3090 and A5000, and it showed: 7.RuntimeError: Given groups=1, weight of size [8, 1, 1, 1], expected input[12, 6, 256, 704] to have 1 channels, but got 6 channels instead , so I setmmdet3d/models/vtransforms/base line37 and line38: False, but the results of the test MAP = 58, and after I used the two other old version to test, the result is only 58, far lower than the final result, is the reason I used one GPU?Here are the results of my single GPU tests in nuscenes-mini: This is my run command: torchpack dist-run -np 1 python tools/test.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml pretrained/bevfusion-det.pth --eval bbox

[gerardmartin2]

@wstzx Did you solve it?

[zhijian-liu]

Thank you for your interest in our project. This repository is no longer actively maintained, so we will be closing this issue. Please refer to the amazing implementation at MMDetection3D. Thank you again!

[YueWangTO]

感谢您的来信，已收到