dumplicate cumsum in motion training - Githubissues

swc-17 / SparseDrive

SparseDrive: End-to-End Autonomous Driving via Sparse Scene Representation

MIT License

184 stars 11 forks source link

dumplicate cumsum in motion training #6

Closed zhangyp15 closed 4 days ago

zhangyp15 commented 4 days ago

Hi, I found there exist two "cumsum" for motion predictions & labels in preparing the motion target sampler (https://github.com/swc-17/SparseDrive/blob/main/projects/mmdet3d_plugin/models/motion/target.py#L28) and before the regression loss (https://github.com/swc-17/SparseDrive/blob/main/projects/mmdet3d_plugin/models/motion/motion_planning_head.py#L398).

Is it as expected ?

zhangyp15 commented 4 days ago

sorry, it was my misunderstanding.