Megvii-BaseDetection / Megvii-BaseDetection/YOLOX

Bad accuracy when training on custom data

Open
#1,766 2 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

Hi, I get bad training results when using YOLOX-s.
I have a custom dataset with 30 classes and approximately 500 images. Each image contains multiple class objects, so the dataset size should be enough. No matter what epoch and batch size I use, COCOAP50 is always around 0.30 and COCOAP50_95 is 0.40.
For the reference, I have used the same dataset to train YOLOv7 and YOLOv8 and the accuracy is close to 90%.

Any suggestions on how to make YOLOX training better? (Apart from the basics - epoch size, batch size, dataset augmentation)

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

Begin by reproducing YOLOX-s training on the reported 30-class, approximately 500-image custom dataset and record COCOAP50 and COCOAP50_95. Compare the training configuration and evaluation results with the reported YOLOv7 and YOLOv8 runs; done requires identifying a reproducible cause or actionable configuration issue and validating improved accuracy.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
computer-vision, machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.