search

skyhehe123 / SA-SSD

SA-SSD: Structure Aware Single-stage 3D Object Detection from Point Cloud (CVPR 2020)
492 stars 106 forks source link

Should val be added to the training when submitted on the official website? #56

Open tyjiang1997 opened 4 years ago

tyjiang1997 commented 4 years ago

Hello, do you need to add the val dataset to the training to submit your results on the KITTI official website? I did so, but the accuracy decreased by 1 percentage points. Can you tell me the reason for this?

aqrliu commented 4 years ago

@LittleYuanzi @skyhehe123 hello,can you get the same mAP as authors?,I clone this project,and download the pretrained model ,but the result mAP is lower about 6％．and I trained the model with cfg (without modified),the mAP is lower still.

tyjiang1997 commented 4 years ago

@LittleYuanzi @skyhehe123 hello,can you get the same mAP as authors?,I clone this project,and download the pretrained model ,but the result mAP is lower about 6％．and I trained the model with cfg (without modified),the mAP is lower still.

No. I can't either.