Hao-Hsiang Yang

TL;DR: A novel two-stage and end-to-end network that comprises two characteristics, the wavelet U-net, which uses discrete wavelet transform and inverse discreteWavelet transform to replace down-sampling and up-sampled, to extract edge features and the chromatic adaptation transform which can be implemented by convolutional layers mathematically to enhance images.

TL;DR: Li et al. as discussed by the authors proposed a lightweight adaptive feature fusion network (LAFFNet), which subsumes multiple branches with different kernel sizes to generate multi-scale feature maps adaptively, and channel attention is used to merge these feature maps.

TL;DR: Y-Net as mentioned in this paper proposes a wavelet structure SIMilarity (W-SSIM) loss function in the training step, which divides the image into different sized patches with different frequencies and scales.

TL;DR: The NTIRE 2021 depth guided image relighting challenge as mentioned in this paper focused on one-to-one relighting where the goal is to transform the illumination setup of an input image (color temperature and light source position) to the target illumination setup.

TL;DR: Experiments on several datasets show that the proposed student model can achieve promising results on different bad weather removal tasks simultaneously, and a novel loss function called multi-contrastive knowledge regularization (MCR) loss is proposed.