deepinsight / deepinsight/insightface
How to set the scale parameter s
- Dominant language
- Python
- Stars
- 29.7k
- Forks
- 6.1k
- PR merge metrics
- No merged PRs in 30d
Description
Hello
How are you?
Thanks for contributing the _Arcface_ paper and this project.
I have a trouble to set the scale parameter s in training an _Arcface_ model.
I used _Resnet50_ and _ms1m_ dataset.
You and many developers use _s_=64.0 but I could not find how to set the scale parameter _s_.
I read a paper for _AdaCos_.
As you know, _AdaCos_ uses _s=sqrt(2)*ln(classes-1)_ as the scale parameter in the fixed parameter mode.
For _ms1m_ dataset (85K classes), _s_ is about 16.06.
I would like to hear your opinion for this.
Thank you.
Contributor guide
No contributing guide indexed for this repository
Research direction
No file, test, or entry point is identified. Start by locating the ArcFace training configuration and reviewing how the scale parameter is currently exposed, then compare the documented fixed value with the AdaCos alternative. Done means documenting the supported setting and its rationale clearly for MS1M and similar class counts.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- computer-vision, machine-learning
- Issue type
- Documentation
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100