Getting stuck at 0% with python eval.py

[carrapatofa]

Hello,

I like your work and would like to use it, but I am having several issues to get it to run. I don't need to train data but want to use your pre-trained model.

I downloaded the "data_set1_5classes" and put it into the PointCloudSegmentation folder.
I downloaded your pretrained model and put the right path in the config file eval.yaml
I changed all the paths as you described in eval.yaml and the data.fold points to the "data_set1_5classes" files.

when then running python eval.py, it seems to load but I am stuck at 0%.. Do you have any idea? Did i miss something in any file to setup?

And as a second thing, i wanted to try out your large_PC_predict.sh (as I have large point clouds to be processed and wanted to try out your tiling), but it seems as it already wants the point clouds (e.g. a point cloud i want to have predicted) to have attributes like semantic_seg and tree_ID, but my raw point clouds (obviously) dont have these attributes. Am I missing something? Do i need to preprocess them before calling the large_PC_predict.sh?

do the point clouds need to be in .ply format already?

I hope I could clarify what i wanted to point out and where I am stuck. please ask if you need anything more from my side.
Thank you very much in advance

[bxiang233]

Hello,

I like your work and would like to use it, but I am having several issues to get it to run. I don't need to train data but want to use your pre-trained model.

I downloaded the "data_set1_5classes" and put it into the PointCloudSegmentation folder. I downloaded your pretrained model and put the right path in the config file eval.yaml I changed all the paths as you described in eval.yaml and the data.fold points to the "data_set1_5classes" files.

when then running python eval.py, it seems to load but I am stuck at 0%.. Do you have any idea? Did i miss something in any file to setup?

And as a second thing, i wanted to try out your large_PC_predict.sh (as I have large point clouds to be processed and wanted to try out your tiling), but it seems as it already wants the point clouds (e.g. a point cloud i want to have predicted) to have attributes like semantic_seg and tree_ID, but my raw point clouds (obviously) dont have these attributes. Am I missing something? Do i need to preprocess them before calling the large_PC_predict.sh?

do the point clouds need to be in .ply format already?

I hope I could clarify what i wanted to point out and where I am stuck. please ask if you need anything more from my side. Thank you very much in advance

Hi, thank you for your interest!

Regarding your first issue, there was a similar case reported in the past:
#1 (comment).
The user encountered the same problem where python eval.py got stuck. They solved it by adjusting their environment setup—specifically, by installing missing dependencies or adjusting package versions.
For your reference:
prs-eth/PanopticSegForLargeScalePointCloud#7 (comment)
and
https://github.com/prs-eth/ForAINet/blob/341e46f16884c521233e4457b72acb60e6824736/PointCloudSegmentation/Dockerfile

Did you follow the environment setup guide here?
https://github.com/prs-eth/PanopticSegForLargeScalePointCloud?tab=readme-ov-file#set-up-environment
The environment setup is the same as ForAINet.

Also, you don’t need to install torch-points3d. The key requirement is to install MinkowskiEngine correctly. torch-points3d includes various backbones that we don’t use. Sorry for the inconvenience, but installing torch-points3d can indeed be tricky. I might not be able to address all installation issues directly, but I can share the setups that worked for me. If you still encounter problems with torch-points3d, I suggest checking its official issues and discussions, as many users have shared their experiences there. Hope this helps, and good luck!

For your second question—yes, you need to have those attributes. However, as mentioned here:
#4 (comment), you can assign dummy values (e.g., 0) if needed.

Regarding the format, yes, the input point clouds need to be in .ply format. You can convert them accordingly.

Hope this helps! Let me know if you have further questions.

Best,
Binbin