Megvii-BaseDetection / Megvii-BaseDetection/YOLOX

validation AP fluctuation and validation loss curve increasing

Open
#1,715 1 comment 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, I am training a custom dataset with YOLOX-nano model. I did not change the nano training parameters except the number of classes which is 1 for my dataset. The Validation AP 0.50:0.95 looks like this:
image

Is it normal to see these fluctuations?

The other issue is calculating validation loss to see if it is overfitting. Is anyone try to log validation loss? I modified the code a bit to calculate validation loss. But somehow it looks weird. It fluctuates a lot and it has an increasing trend.
image

And this is total training loss by the way:
image

What is going wrong? Does anyone have any idea to fix this?
Thanks.

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 reviewing the YOLOX-nano custom-dataset training setup and the modified validation-loss calculation described in the issue. No source file or test is named, so reproducing the AP and validation-loss behavior would be needed before defining a fix; done would mean explaining or correcting the reported fluctuation and increasing validation-loss trend.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
15/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.