Can't run the static scene grasping block on custom data

[PannagaS]

Hi, first of all, this is a phenomenal paper and very well implemented. I was trying to implement this on the given example_data, which worked fine without any errors.

However, when I try to run the entire implementation on my custom data set, I can't run the static scene grasping command - python realbot_ui.py -m outputs/path_to_my_data

So, I started debugging and found that the keys of feat_decoder.pth is NOT SAME for the example data and for my custom data. I printed the keys of feat_decoder.pth for both the datasets;

For example_data, I am getting -

And, for my custom_data, I am getting -

As you can see, in the feat_decoder.pth that is generated for my custom_data, I am NOT having 'part_conv1.bias', 'part_conv1.weight_g', 'part_conv1.weight_v' keys.

This is causing an issue when I run - python realbot_ui.py -m outputs/path_to_my_data. How do I fix this? Any help to resolve this issue is deeply appreciated.

PS : In case the screenshot of the terminal is not super clear, I also wrote it down for reference -

Thank you!

[jimazeyu]

I think you forgot to add "--feature_type "clip_part" when computing the features: _python feature-splatting-inria/train.py -s scene_data/example_data -m outputs/example_data --iterations 3000 --feature_type "clippart"

[PannagaS]

Yes, I realize that now. Thank you very much @jimazeyu .