Single image dehazing via cycle-consistentadversarial networks with a multi-scale hybridencoder-decoder and global correlation loss
we propose novel cycle-consistent adversarial networks with a multi-scale hybrid encoder-decoder and global correlation loss for single image dehazing task. The requirement of paired training data is eliminated by combining two generators and discriminators into a cycle-consistent adversarial network.
*Fig. 1. Comparison of qualitative results on the NYU-Depth dataset.
*Fig. 2. Comparison of qualitative results on the Middlebury dataset.
*Fig. 3. Comparison of qualitative results on the Indoor SOTS dataset.
*Fig. 4. Comparison of qualitative results on the O-HAZE dataset.
*Fig. 5. Comparison of qualitative results on the Sea-fog dataset.
*Fig. 6. Comparison of qualitative results on the real-world images.
torch>=0.4.1 torchvision>=0.2.1 dominate>=2.3.1 visdom>=0.1.8.3
If you find this code useful, please cite:
Yao et al., Single image dehazing via cycle-consistentadversarial networks with a multi-scale hybridencoder-decoder and global correlation loss.