support (Statistically unbiased prediction enables accurate denoising of voltage imaging data) github(https://github.com/NICALab/SUPPORT) 왼쪽이 noise image 픽셀 히스토그램, 오른쪽이 support 실행후 픽셀 히스토그램
rank 1 KBNe t29.54 KBNet: Kernel Basis Network for Image Restoration
- 1.학습된 pretrained_model을 다운 받음(파일위치:https://mycuhk-my.sharepoint.com/personal/1155135732_link_cuhk_edu_hk/_layouts/15/onedrive.aspx?id=%2Fpersonal%2F1155135732%5Flink%5Fcuhk%5Fedu%5Fhk%2FDocuments%2Fshare%2FKBNet&ga=1)
- 2.gau_gray_15.pth,gau_gray_25.pth,gau_gray_50.pth 중 우리에게 잘 맞는 것은 gau_gray_25.pth로 판단
-
- denoising 한장당 6분정도 (gpu 3090 24gb)
-
- denoising 한 이미지를 teed로 boundary detection ->결과: 제일 좋은 성능 보여줌