search

qq456cvb / CanonicalVoting

Canonical Voting: Towards Robust Oriented Bounding Box Detection in 3D Scenes (CVPR2022)
46 stars 7 forks source link

Training & Evaluating model on Sunrgbd #5

Open guthasaibharathchandra opened 1 year ago

guthasaibharathchandra commented 1 year ago

Hi could you please leave instructions on how to train and evaluate your model on sunrgbd dataset. It seems like for sunrgbd, one needs to train/evaluate it using mmdetection3d framework? or could I directly use the scripts that you have provided? Could you please clarify or let me know if i'm missing something here? Thanks!

qq456cvb commented 1 year ago

We first train a separate Canonical Voting module to generate the center heatmaps. Then this center heatmap is leveraged as a custom sampler and then we use the mmdetection3d framework to refine these proposals. For the first training phase, the dataloader is quite similar to that of ScanNet (details given in the paper); while for the second training phase, you can directly use brnetcanon.py to train and evaluate with mmdetection3d.