search

ZrrSkywalker / Point-NN

[CVPR 2023] Parameter is Not All You Need: Starting from Non-Parametric Networks for 3D Point Cloud Analysis
MIT License
477 stars 50 forks source link

Making own dataset #10

Open Cipher-zzz opened 1 year ago

Cipher-zzz commented 1 year ago

I am trying to test this Non-parametric Network with some outdoor 3D scenes, including NPM3D, Kitti and data captured by ourselves. If you are also testing that, when would you expect to update the result and code in this repo?

I noticed a 'data_scan.py' file for processing the Scannet dataset exists, but no corresponding 'run_nn_scan.py'. If you have the code for running the Scannet with Non-parametric Network and can upload it, it may help me a lot! Or could you provide some instructions about preparing other datasets, especially on arranging the Point-Memory Bank?

Many thanks.

ZrrSkywalker commented 1 year ago

Thanks for your interest. Sorry for misleading. data_scan.py refers to the ScanObjectNN dataset for shape classification. We haven't try Point-NN on ScanNet dataset and would leave it as a future work.

For point-memory bank for scene-level dataset, I suppose it's better to refer to the bank of Point-NN for part segmentation, which we have open-sourced at here. The feature memory is better to be clustered by some rules to save GPU memory and obtain prototypical characters.

Cipher-zzz commented 1 year ago

Thanks for your reply.