diff --git a/README.md b/README.md index d01704e317116567e749564cc2a3056f20d288cb..71a8f205cc4c5bd8ffaeb56f478d415d38f61b1d 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,10 @@ **About PyTorch 1.6.0** * Now the main branch supports PyTorch 1.6.0 by default. -# Attention transfer을 활용한 EDSR-PyTorch모델 경량화 +# knowledge distillation을 활용한 EDSR-PyTorch모델 경량화 **About PyTorch 1.6.0** - * feature map transfer 방식을 위에서 언급한 EDSR에 적용하여 모델 경량화 실험 진행 예정 + * feature map transfer 방식 등, KD에서 사용하는 방식으로 EDSR에 적용하여 모델 경량화 실험 진행 예정  @@ -17,6 +17,14 @@ This repository는 **"Enhanced Deep Residual Networks for Single Image Super-Res 중간 실험 결과 AT의 하이퍼파라미터를 1.0으로 했을 때, 결과. 오히려 지식 증류를 받은게 더 성능이 떨어짐. +<img width="670" alt="스크린샷 2023-05-22 오후 9 53 37" src="https://github.com/iyj0121/Junior-Project/assets/90498398/d9865cb6-7d21-4e6d-8af5-2bba43474837"> + +<img width="670" alt="스크린샷 2023-05-18 오전 1 35 53" src="https://github.com/iyj0121/Junior-Project/assets/90498398/00e1da90-3713-4b62-b060-4b54dc1bb8ec"> + +<img width="681" alt="스크린샷 2023-05-18 오전 1 35 35" src="https://github.com/iyj0121/Junior-Project/assets/90498398/bd3eb6c3-65c1-4db3-aeaa-2482ef92508f"> + +KD방식이 아닌 pruning 등 다른 방식으로 모델 경량화를 이어나갈 예정. + [1] Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee, **"Enhanced Deep Residual Networks for Single Image Super-Resolution,"** <i>2nd NTIRE: New Trends in Image Restoration and Enhancement workshop and challenge on image super-resolution in conjunction with **CVPR 2017**. </i> [[PDF](http://openaccess.thecvf.com/content_cvpr_2017_workshops/w12/papers/Lim_Enhanced_Deep_Residual_CVPR_2017_paper.pdf)] [[arXiv](https://arxiv.org/abs/1707.02921)] [[Slide](https://cv.snu.ac.kr/research/EDSR/Presentation_v3(release).pptx)] ``` @InProceedings{Lim_2017_CVPR_Workshops,