demo_pipeline.sh hangs at sfm reconstruction

[apirrone]

Hello, when I run bash scripts/demo_pipeline.sh $OBJ_NAME, the process seems to hang when performing the keypont-free sfm reconstruction

I have this :

2023-02-28 12:02:06.151 | INFO     | __main__:sfm_core:193 - Keypoint-Free SfM coarse reconstruction begin...
                                                                                                                             (match_worker_ray_wrapper pid=26954) Global seed set to 666                                            | 0/725 [00:00<?, ?it/s]
(match_worker_ray_wrapper pid=26955) Global seed set to 666
(match_worker_ray_wrapper pid=26953) Global seed set to 666
(match_worker_ray_wrapper pid=27030) Global seed set to 666

Then nothing happens, and the process does not seem to be doing anything

Any idea what's going on ?

Thanks !

Edit: I should add that I am trying to run the demo on the povided demo_cam.zip data

[MurphyLe]

First, use the command：“nvidia-smi” to check whether the graphics card works properly！
If the graphics card does not work，Please comment out the following code：

src/KeypointFreeSfM/coarse_match/coarse_match.py
Line 63
#pb.print_until_done() if pb is not None else None

For me, this line of code blocks my process。

[apirrone]

Thanks @MurphyLe for the advice.

My GPU is working properly however (I spend quite some time training deep learning models on it :) )

[Knightbreeze]

Hello, when I run bash scripts/demo_pipeline.sh $OBJ_NAME or python run.py +preprocess=sfm_train_data.yaml use_local_ray=True,I have encountered the following problems in the process of keypont-free sfm reconstruction.

File "/home/xxx/deeplearning/onepose++/OnePose++/src/models/OnePosePlus/backbone/init.py", line 8, in build_backbone
if config['type'] == 'ResNetFPN':
KeyError: 'type'

I don't know where the problem is. It took me a long time to solve it. I don't know if you have encountered this problem.
I hope I can get your reply. thank you.@MurphyLe @apirrone

[hxy-123]

Hello, when I run bash scripts/demo_pipeline.sh $OBJ_NAME, the process seems to hang when performing the keypont-free sfm reconstruction

I have this :

2023-02-28 12:02:06.151 | INFO     | __main__:sfm_core:193 - Keypoint-Free SfM coarse reconstruction begin...
                                                                                                                             (match_worker_ray_wrapper pid=26954) Global seed set to 666                                            | 0/725 [00:00<?, ?it/s]
(match_worker_ray_wrapper pid=26955) Global seed set to 666
(match_worker_ray_wrapper pid=26953) Global seed set to 666
(match_worker_ray_wrapper pid=27030) Global seed set to 666

Then nothing happens, and the process does not seem to be doing anything

Any idea what's going on ?

Thanks !

Edit: I should add that I am trying to run the demo on the povided demo_cam.zip data

Hi, sorry for the late reply! It seems that the problem occurs during the SfM phase. You can try to close the parallelized matching by setting use_local_ray=False: python run.py +preprocess=sfm_train_data.yaml use_local_ray=True. The errors will thus be reported. Note that I think the problem may be due to the submodule LoFTR is not properly installed. You can install it by running git submodule update --init --recursive. Another problem may be due to the CUDA OOM caused by parallelized matching.

[hxy-123]

Hello, when I run bash scripts/demo_pipeline.sh $OBJ_NAME or python run.py +preprocess=sfm_train_data.yaml use_local_ray=True,I have encountered the following problems in the process of keypont-free sfm reconstruction.

File "/home/xxx/deeplearning/onepose++/OnePose++/src/models/OnePosePlus/backbone/init.py", line 8, in build_backbone
if config['type'] == 'ResNetFPN':
KeyError: 'type'

I don't know where the problem is. It took me a long time to solve it. I don't know if you have encountered this problem. I hope I can get your reply. thank you.@MurphyLe @apirrone

Hi, I think the problem may be due to the submodule LoFTR is not properly installed. You can install it by running git submodule update --init --recursive.

[Knightbreeze]

Hello, when I run bash scripts/demo_pipeline.sh $OBJ_NAME or python run.py +preprocess=sfm_train_data.yaml use_local_ray=True,I have encountered the following problems in the process of keypont-free sfm reconstruction.

File "/home/xxx/deeplearning/onepose++/OnePose++/src/models/OnePosePlus/backbone/init.py", line 8, in build_backbone
if config['type'] == 'ResNetFPN':
KeyError: 'type'

I don't know where the problem is. It took me a long time to solve it. I don't know if you have encountered this problem. I hope I can get your reply. thank you.@MurphyLe @apirrone

Hi, I think the problem may be due to the submodule LoFTR is not properly installed. You can install it by running git submodule update --init --recursive.

Thank you very much for your reply. It is true that the installation of LoFTR was not installed correctly, but the git submodule update --init --recursive command will not work correctly on my computer. I have completed the git configuration, and then I resolved the problem by git clone separately. And I do need to change use_local_ray to False on my computer. Thanks again for your reply!

[xiexh20]

Hi, I think the problem may be due to the submodule LoFTR is not properly installed. You can install it by running git submodule update --init --recursive.

Thank you very much for your reply. It is true that the installation of LoFTR was not installed correctly, but the git submodule update --init --recursive command will not work correctly on my computer. I have completed the git configuration, and then I resolved the problem by git clone separately. And I do need to change use_local_ray to False on my computer. Thanks again for your reply!

Hi, did you just git clone the latest LoFTR repo? I did the same, but still the LoFTR module failed to find matches and it even fails to run for the demo data. Do you have any thoughts? thanks!