VDSR was proposed by Jiwon Kim et al. in 2016. The author mainly uses a deep convolutional network based on VGG-Net, which only learns residuals and use extremely high learning rates (104 times higher than SRCNN) enabled by adjustable gradient clipping, and ultimately has a great advantage in image quality performance.
| Model | Download | PSNR (dB) |
|---|---|---|
| VDSR | model weight | 25.18~37.53 |
- Accurate Image Super-Resolution Using Very Deep Convolutional Networks
- BobLiu20/SuperResolution_Caffe
- Web
NO LICENSE