Apply amp to object detection
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 9k
- Forks
- 1.5k
- Avg merge
- 2d 4h
- Merged PRs (30d)
- 3
Description
Hi, I could not find any detail example about applying AMP to object detection models, e.g. Faster-RCNN, could I ask where I can find the example?
I have also tried to implemente some models, but in object detection the input size is usually not fixed and cudnn.benchmark have to be set as false, but https://github.com/NVIDIA/apex/blob/a8c2b7dd12cd8dc3ceb7137581dd1af90f5ee9d5/examples/imagenet/main_amp.py#L75 is set to be true. Is it essential to FP16 training?
I am also curious about the speed and memory gain when switching FP32 to FP16, can it achieve the same gain as image classification?
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
Start by reading examples/imagenet/main_amp.py around line 75 and reviewing the existing AMP examples for guidance on object detection. Document or add a Faster-RCNN-style object detection example, clarify whether cudnn.benchmark affects AMP training, and report expected speed and memory differences between FP32 and FP16.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- documentation, machine-learning
- Issue type
- Documentation
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100