Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: Training with MMPose #25

Open
KidsareBornStars opened this issue Jan 31, 2023 · 5 comments
Open

feat: Training with MMPose #25

KidsareBornStars opened this issue Jan 31, 2023 · 5 comments
Assignees

Comments

@KidsareBornStars
Copy link
Contributor

KidsareBornStars commented Jan 31, 2023

Description

  • MMPose Training Status

Details

  • LiteHRNet18 / HRNet32 / HRNet48 / MobileNetV2 / ResNet50 기반 모델들 학습 비교

Reference / Image

@KidsareBornStars KidsareBornStars self-assigned this Jan 31, 2023
@KidsareBornStars
Copy link
Contributor Author

10000장으로 학습했을 때 AP는 잘 나오지만, 실제로 test data를 넣어줬을 때 잘 되지 않는 현상 발견
-> 따라서 20000+장으로 재학습

@KidsareBornStars
Copy link
Contributor Author

5초영상 기준
hrnet_w32의 경우 inference 15초

@KidsareBornStars
Copy link
Contributor Author

면접 영상에 Inference시 전혀 되지 않는 문제가 발생하여
기존 pretrained된 모델에 finetuning하는 방법을 사용.

@KidsareBornStars
Copy link
Contributor Author

20000장으로 Finetuning한 모델의 성능이 좋지 않아(특히 손을 감지하기 힘들어함)
데이터셋 중 (Left/Right 제외하고) 면접 Test Data에 특화된 Up/Down/Front각도에서 찍은
손이 움직이는 20~110프레임까지 15000장으로 finetuning

@KidsareBornStars
Copy link
Contributor Author

From Scratch로 사용하기 위해 GCP에 GPU탑재하여 VM생성하였으나, 학습 코드를 실행하면 서버가 터져버리는 문제 발생

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant