Jiao J, Tu W C, He S, et al. Formresnet: Formatted residual learning for image restoration[C]//Computer Vision and Pattern Recognition Workshops (CVPRW), 2017 IEEE Conference on. IEEE, 2017: 1034-1042.
1. Overview
1.1. Motivation
- directly learn the clean images may suffer gradient problem
- directly learn the high-frequency residual may harm the structure detials
- L2 loss suffers from blur
In this paper, it
- proposed residual formating layer to recover the latent clean image
- learn the structure detial
- proposed cross-level loss net
- Experiments on denoising, SR, de-raining, inpainting, enhancement
2. Methods
2.1. Residual Formatting Layer
- aim to reduce the corrupation on the input image
- layer can be stacked
2.2. Cross-Level Loss Net
2.2.1. Pixel Loss
2.2.2. Feature Loss
2.2.3. Gradient Loss
- sobel
2.2.4. Loss Function
3. Experiments
3.1. Comparison
3.2. Time
