Augmix 실험결과 #40
sne12345
started this conversation in
Show and tell
Replies: 1 comment
-
진짜 inference 결과와는 다르게 LB가 높아지네요... |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
What is Augmix? and parameters
parameters
Augmix 효과 based on paper
-> Mixing augmentations allows us to generate diverse transformations, which are important for inducing robustness as a common failure mode of deep models in the arena of corruption robustness is the memorization of fixed augmentations
train 데이터 확인
[비교군] 아무런 옵션도 주지 않은 기본 HRNet
[Augmix] severity=3, width=3, depth=-1(random), alpha=1., p=1
[Augmix] severity=3, width=7, depth=-1(random), alpha=1., p=1
[Augmix] severity=3, width=14, depth=-1(random), alpha=1., p=1 / 논문 참고해서 RandomCrop(width=256, height=256), HorizontalFlip(), ColorJitter() 추가
[Augmix] severity=3, width=3, depth=-1(random), alpha=1., p=1 / 논문 참고해서 contrast, color, brightness, sharpness 제외
test 데이터 확인 & 실험결과
[비교군] 아무런 옵션도 주지 않은 기본 HRNet
[Augmix] severity=3, width=3, depth=-1(random), alpha=1., p=1
LB 0.522
클래스 전체적으로 valid IoU 향상
[개인적인 의견] 경계를 깔끔하게 잘 예측하는듯
-> test data 시각화에서 HRNet 모델이 짤려서 예측한 마스크가 많아 보였는데
-> Augmix를 통해 각도를 다르게 여러 이미지를 합쳐서 좀 더 경계를 일반화시킨게 아닐까?
[Augmix] severity=3, width=7, depth=-1(random), alpha=1., p=1
[Augmix] severity=3, width=14, depth=-1(random), alpha=1., p=1 / 논문 참고해서 RandomCrop(width=256, height=256), HorizontalFlip(), ColorJitter() 추가
[Augmix] severity=3, width=3, depth=-1(random), alpha=1., p=1 / 논문 참고해서 contrast, color, brightness, sharpness 제외
..돌리는중...
Beta Was this translation helpful? Give feedback.
All reactions