Wrong 'img_norm_cfg' setting

hustvl / VAD

[ICCV 2023] VAD: Vectorized Scene Representation for Efficient Autonomous Driving

https://arxiv.org/abs/2303.12077

Apache License 2.0

714 stars 78 forks source link

Closed zhiqi-li closed 1 year ago

zhiqi-li commented 1 year ago

It seems that you used the wrong img_norm_cfg for the resnet50 (pytorch format).

It should be

img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)

rb93dett commented 1 year ago

Thank you for pointing out : ), we have updated the configs to use the right img_norm_cfg.

tyzhi commented 1 year ago

I noticed that the img_norm_cfg is the same to that of UniAD . Could you please explain whether it will affect model's performance?