DACON Monthly the 2nd Computer Vision Learning Competition

Algorithm | CV | Classification | Control | Accuracy

  • moneyIcon Prize : 540만원
  • 45 Users Completed
Closed

 

중간평가 Private 1등, Efficient b8 model, albumentation 데이터 증강

2022.09.29 20:31 609 Views language

학습을 진행하면서 수동적으로 LR 수정하면서 진행했습니다.
val_acc가 잘 나오는 batch_size를 찾아보려 여러번 수정하며 돌렸습니다.
seed를 고정시킨다고 해도 여러 TPU, 다른 서버 환경문제로 정확히 kfold를 진행할 수 없을 거 같아  <-- 확실하지 않습니다.
랜덤 array를 생성시킨 후 pickle로 저장하여 불러와서 train, val index를 조절하는 식으로 학습했습니다.

Code