search

Tsinghua-MARS-Lab / Occ3D

MIT License
395 stars 25 forks source link

Why the performance for BevFormer in paper is higher than the baseline in cvpr challenge page? #3

Closed promesse closed 1 year ago

promesse commented 1 year ago

image image

page:https://github.com/CVPR2023-3D-Occupancy-Prediction/CVPR2023-3D-Occupancy-Prediction/blob/main/docs/getting_started.md

waveleaf27 commented 1 year ago

Hi, the main reason is adopting more effective training strategies, such as adjusting loss weight and using data augmentation.

promesse commented 1 year ago

Hi, the main reason is adopting more effective training strategies, such as adjusting loss weight and using data augmentation. Did you use camera_maks during training?

waveleaf27 commented 1 year ago

We not use camera_mask for the metrics you mentioned.

promesse commented 1 year ago

Even for loss？

waveleaf27 commented 1 year ago

when adopting camera_mask, the metrics may improve significantly. You can ref BevDet for more detail.