about the given results of the model detection of upidet_release.pth

[thelandord]

I tested the pre-trained model under pytorch1.10.0, cuda11.3,python=3.8, and the result is different from the published result, is it because of the version of pytorch?

[Eaphan]

The reason for this discrepancy could be due to differences in library versions and the presence of noise introduced during pre-processing.

[thelandord]

The reason for this discrepancy could be due to differences in library versions and the presence of noise introduced during pre-processing.

Thank you for your help!
I have another problem, I encountered the following bug while training, at line 161 of dataset.py
data_dict['segmentation'] = [data_dict['segmentation'][ind] for ind in selected]
There is a bug: IndexError: list index out of range.
I print it and get the output:

Why is the len of data_dict['segmentation'] less than the len of selected? Do I need to align the selected filter to data_dict['segmentation']?

[Eaphan]

Please follow the README.md to prepare data and train.

[thelandord]

Please follow the README.md to prepare data and train.

thanks for your reply！ I have replaced the .pkl file.I meet the bug after replaced pkl. But the len of data_dict['segmentation'] less than the len of selected.Is there a detail I'm missing?

[Eaphan]

Have you replaced all original pkl files, including the kitti_dbinfos_train.pkl.

[thelandord]

Have you replaced all original pkl files, including the kitti_dbinfos_train.pkl.

yes, i have replaced the kitti_dbinfos_train.pkl, kitti_infos_train.pkl ,kitti_infos_val.pkl. i print the pkl as the following pictures

[Eaphan]

Could you use the command "git diff" to see what changes you have made?