This is the PyTorch implementation for our CVPR'21 paper. The model can removal hazy, smoke or even water impurities.
The repository includes:
- Source code of our model.
- Training code for O-hazy dataset.
- Testing code for O-hazy dataset
- Pre-trained model for O-hazy dataset.
Setup: 依赖的库 torch, numpy, tqdm, torchvision, kornia, opencv-python
Training 将带雾训练数据集放在./hazy 文件夹下 对应的清晰数据集放在./gt文件夹下。 运行命令 python train.py。 训练过程可在./result文件夹下找到。 模型保存在./model文件夹下。
Test model 将需要测试的数据集放在./OHAZE_test文件下。 运行命令 python test_model.py。 测试结果可在./test_result文件夹下找到。
Requirements： Python 3.7 PyTorch 1.6.0 CUDA 10.0 Ubuntu 16.04
Dataset (Daytime) Link：https://pan.baidu.com/s/1sqJpxvt1-ONqcuG7RLTq0A Password：vodp
Nighttime with Haze 4K dataset Link： https://pan.baidu.com/s/1pxmsFOU-3ELNgR8KtvWW1g Password：5h6A
Pre-model Link：https://pan.baidu.com/s/1UwDL8rzTVFYFDwOsU9TBlA Password：u5pk
4K real-world video with hazy Link：https://pan.baidu.com/s/1wqQKEPLnzTPqANAr-D5Mxw Password：p83y
4K real-world images Link：https://pan.baidu.com/s/1u5Wq3aKBVxZhAkVdHsdMVQ Password：776z
4K nighttime videos Link：https://pan.baidu.com/s/1lCPh70aKfOd2Zg6Ah37jGg Password：h1in
Cite: { title = {Ultra-High-Definition Image Dehazing via Multi-Guided Bilateral Learning}, booktitle = {CVPR}, year = {2021} }
Single UHD Image Dehazing via Interpretable Pyramid Network
Currently, most single image dehazing models cannot run an ultra-high-resolution (UHD) image with a single GPU shader in real-time. To address the problem, we introduce the principle of infinite approximation of Taylor's theorem with the Laplace pyramid pattern to build a model which is capable of handling 4K hazy images in real-time. The N branch networks of the pyramid network correspond to the N constraint terms in Taylor's theorem. Low-order polynomials reconstruct the low-frequency information of the image (e.g. color, illumination). High-order polynomials regress the high-frequency information of the image (e.g. texture). In addition, we propose a Tucker reconstruction-based regularization term that acts on each branch network of the pyramid model. It further constrains the generation of anomalous signals in the feature space. Extensive experimental results demonstrate that our approach can not only run 4K images with haze in real-time on a single GPU (80FPS) but also has unparalleled interpretability. The developed method achieves state-of-the-art (SOTA) performance on two benchmarks (O/I-HAZE) and our updated 4KID dataset while providing the reliable groundwork for subsequent optimization schemes.