search

CVMI-Lab / PLA

(CVPR 2023) PLA: Language-Driven Open-Vocabulary 3D Scene Understanding & (CVPR2024) RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding
Apache License 2.0
260 stars 11 forks source link

Bad results when running inference of annotation-free models #52

Open argosdh opened 3 months ago

argosdh commented 3 months ago

Hi, thanks for your great work!

I'm encountering some problem when running the annotation-free setup models e.g. [RegionPLC + OpenScene + SparseUNet32] or [RegionPLC + SparseUNet32]. When inferenced on scannet validation, every point would be given the same label. This is my scripts used for running: python3 test.py --cfg_file cfgs/scannet_models/spconv_clip_openvocab_test.yaml --ckpt ../checkpoints/sparseunet32_596.pth --save_results semantic

I barely changed the yaml file, only for those categories I want to add, like from CATEGORY_NAMES: [door, window, desk, keyboard, others] to CATEGORY_NAMES: [trash can, television ,door, window, desk, keyboard, chair, others]

this is my yaml file, since I only need the qualitiative results, I put all my labels in base labels. Selection_720

Thank you if you can reply !

jihanyang commented 3 months ago

Hello, you used the wrong config file here, you can follow the config file indicated in https://github.com/CVMI-Lab/PLA/blob/regionplc/docs/MODEL.md. Such as https://github.com/CVMI-Lab/PLA/blob/regionplc/tools/cfgs/scannet200_models/zs/spconv_clip_caption_openscene.yaml or https://github.com/CVMI-Lab/PLA/blob/regionplc/tools/cfgs/scannet200_models/zs/spconv_clip_caption.yaml

argosdh commented 3 months ago

Hello, you used the wrong config file here, you can follow the config file indicated in https://github.com/CVMI-Lab/PLA/blob/regionplc/docs/MODEL.md. Such as https://github.com/CVMI-Lab/PLA/blob/regionplc/tools/cfgs/scannet200_models/zs/spconv_clip_caption_openscene.yaml or https://github.com/CVMI-Lab/PLA/blob/regionplc/tools/cfgs/scannet200_models/zs/spconv_clip_caption.yaml

Thanks, I guess the my repo was the old one before the latest update. I'll try now.