MegCupDenoise-Top6 [link]
- Simple U-net architecture with one downsample
- DAB block includes one Channel Attetion and One Spatial Attention
- There is a DAB block in each encoder stage
- Bayer Image as model's input directly
- Classic image augmentation works