search

megvii-research / PETR

[ECCV2022] PETR: Position Embedding Transformation for Multi-View 3D Object Detection & [ICCV2023] PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images
Other
823 stars 125 forks source link

How to determine the ''resize_lim'' in the ''ida_aug_conf'' according to different input resolutions? #109

Closed xXuHaiyang closed 1 year ago

xXuHaiyang commented 1 year ago

Hello! A very good work!

I've mentioned that in petr_vovnet_gridmask_p4_800x320.py, we have ida_aug_conf = { "resize_lim": (0.47, 0.625), "final_dim": (320, 800), "bot_pct_lim": (0.0, 0.0), "rot_lim": (0.0, 0.0), "H": 900, "W": 1600, "rand_flip": True, }. In petr_r50dcn_gridmask_p4.py, we have ida_aug_conf = { "resize_lim": (0.8, 1.0), "final_dim": (512, 1408), "bot_pct_lim": (0.0, 0.0), "rot_lim": (0.0, 0.0), "H": 900, "W": 1600, "rand_flip": True, }. I wonder that how do we determined the ''resize lim'' according to different input resolution? Is there any recipe? e.g., in the paper we have results of resolution at 384x1056, then in this case what's the ''resize lim''?

Thanks!

yingfei1016 commented 1 year ago

Hi,

(704x256), (0.386,0.55) (800x320), (0.47,0.625) (1024x384), (0.58,0.83) (1408x512), (0.8,1.0) (1600x640), (0.94,1.25)

Some parameters are searched by experiments (1408x512), so they seem irregular. For other input resolution, you can compute the ''resize lim'' base on (1600x640). For example, the ''resize lim'' is (0.75, 1.0) for (1280x512).