Megvii-BaseDetection / Megvii-BaseDetection/YOLOX
Training on very large dataset
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 10.6k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
Hi, my very large training annotation file in COCO format is approximately 60GB on disk. Do you have any tips on how to train with YOLOX on a dataset this large? Problems I've encountered are:
- distributed data parallel tries to load the entire file in each process upfront so I run out of RAM (400GB on my machine)
- A single epoch takes far too long so it seems better to measure and schedule via total iterations instead of epochs
Any help would be greatly appreciated!
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
The issue names no files, tests, or entry points. Start by tracing YOLOX's COCO dataset loading under distributed data parallel and its epoch/iteration scheduling; done would require a confirmed approach for training the 60GB dataset without exhausting RAM and for measuring progress by total iterations.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- data, distributed-systems, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100