Megvii-BaseDetection / Megvii-BaseDetection/YOLOX

Objectness calculation and balancing factor w.r.t reg_loss and class_loss

Open
#1,788 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
10.6k
Forks
2.5k
PR merge metrics
No merged PRs in 30d

Description

Hello, since the Objectness loss (IoU in the case of YOLOX) should take into account all the predictions, not just the positives found with simOTA, how is it weighted the fact that uses all 13x13 + 26x26 + 52x52 predictions while reg_loss and class_loss use just positives ?
I can't even understand if it uses just the positives and their the IoU with respect to the assigned GT or if it uses also the negatives, how the target for the negative is calculated ? Thanks in advance for every answer.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by locating the YOLOX loss and simOTA implementation, then trace how objectness, reg_loss, and class_loss select predictions and construct targets. Compare positive and negative predictions across the 13x13, 26x26, and 52x52 grids. Done means documenting how negative targets are calculated and how objectness is weighted relative to the other losses.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
computer-vision, machine-learning
Issue type
Documentation
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.