search

OpenDriveLab / UniAD

[CVPR'23 Best Paper Award] Planning-oriented Autonomous Driving
Apache License 2.0
3.11k stars 335 forks source link

Question about eval results #157

Closed godkaieethu closed 6 months ago

godkaieethu commented 6 months ago

Thank you for your excellent work.

I follow the steps in train_eval.md

for eval examples(using ckpt this work provided),i got the correct result Amota 0.390. 评估· But after i trained the stage one model and used the ckpt generated to eval, i only got amota 0.373.Is any thing wrong? 训练-评估-stage1

godkaieethu commented 6 months ago

my train process is followed this command: ./tools/uniad_dist_train.sh ./projects/configs/stage1_track_map/base_track_map.py 8 on 8 * A100 80G GPUS

ilnehc commented 6 months ago

@godkaieethu Probably due to GPU numbers. See log here: https://github.com/OpenDriveLab/UniAD/issues/21#issuecomment-1586549953.

godkaieethu commented 6 months ago

@godkaieethu Probably due to GPU numbers. See log here: #21 (comment).

thank you for your reply. According to your log in https://github.com/OpenDriveLab/UniAD/issues/21#issuecomment-1586549953. ,was the 0.393 result got on 16 GPUS? I did my train on 8 gpus ,maybe it is the reason for the difference

ilnehc commented 6 months ago

@godkaieethu Yes. This will make a difference.