autonomousvision / good

[ICLR'23] GOOD: Exploring Geometric Cues for Detecting Objects in an Open World
MIT License
39 stars 8 forks source link

[Reimplementation] #2

Open luoh226 opened 1 year ago

luoh226 commented 1 year ago

Prerequisite

💬 Describe the reimplementation questions

Thanks for your preferct work. I have some confuse when reimplemented your work.

  1. Why img_norm_cfg in config file "phase1_depth.py" different to "phase2_good.py" and "oln_box.py"

    # phase1_depth.py
    img_norm_cfg = dict(
    mean=[101.5219, 159.6869,  97.8256], std=[53.7791, 52.5693, 34.5755], to_rgb=False)
    # phase2_good.py and oln_box.py
    img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)

    How to obtain the regularization parameters dict(mean=[101.5219, 159.6869, 97.8256], std=[53.7791, 52.5693, 34.5755], to_rgb=False) in "phase1_depth.py"?

  2. The starting lr in paper is 0.01 but in config file is 0.02, which one should I choose?

Environment

The same to your work.

Expected results

No response

Additional information

No response