tip
project 준비 방식
model(paperswithcode)
EfficientNet →
nfnet →
timm (
https://github.com/rwightman/pytorch-image-models/tree/master/timm
) →
ViT →
data → 데이터 불균형 문제를 어떻게 처리해야할지, mask dataset훨씬 많아서
https://aihub.or.kr/aidata/27716
TTA :
https://inspaceai.github.io/2019/12/20/Test_Time_Augmentation_Review/
data augmentation (
https://arxiv.org/pdf/2009.08369.pdf
)
Fast augmentation (
https://hoya012.github.io/blog/albumentation_tutorial/
)
💡 idea
GPU도 좋으니 image size를 크게 가져가자 → 512x512(요즘 대세)
Model 🦆