autonomousvision / good

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

[Reimplementation] some question about reimplementation #5

Open Daniellli opened 9 months ago

Daniellli commented 9 months ago

Prerequisite

💬 Describe the reimplementation questions

Thank you for your contribution. How can I modify the code to produce the $AR_A$? I evaluated the VOC-NonVOC model you provided and received the following results:

Screenshot2023-12-24 09 51 14

It seems there are no results about $AR_A$?

Environment

None

Expected results

$AR_A$

Additional information

None

Sino0904 commented 4 months ago

@Daniellli how were you able to replicate the final performance? When I tried it myself from scratch, I got AR@100 = 38.8%, which is lower than what is reported in the paper.