fudan-zvg / DeepInteraction

[NeurIPS 2022] DeepInteraction: 3D Object Detection via Modality Interaction
MIT License
221 stars 16 forks source link

TTA and model ensembling #30

Open NoOneUST opened 1 year ago

NoOneUST commented 1 year ago

Thank you for your great work. Did you provide any tutorial or code on how to do TTA and model ensembling? It seems that these techniques benefit the effectiveness a lot.

Alexander0Yang commented 1 year ago

Hi, @NoOneUST. mmdetection3d has already implement scaling and flipping tests. As for ensemble, please refer to https://github.com/tianweiy/CenterPoint/issues/125#issuecomment-821251419.

SISTMrL commented 1 year ago

Hi, @NoOneUST. mmdetection3d has already implement scaling and flipping tests. As for ensemble, please refer to tianweiy/CenterPoint#125 (comment).

could you please release the config of deepinteraction-large, the repo you released is the deepinteraction-base