Implementation of WResVLM, from the following paper:
Towards Real-World Adverse Weather Image Restoration: Enhancing Clearness and Semantics with Vision-Language Models (ECCV 2024)
Jiaqi Xu, Mengyang Wu, Xiaowei Hu, Chi-Wing Fu, Qi Dou, Pheng-Ann Heng
We use several (pseudo-)synthetic datasets, including Outdoor-Rain, RainDrop, SPA, OTS, Snow100K. Meanwhile, we use real-world data from URHI and our collected real rain and snow images for model training. The real rain and snow images used for training can be downloaded here or here. The 2,320 filtered real rain images from DDN-SIRR and Real3000, used for testing, can be downloaded here or here. The Snow100K-realistic dataset can be downloaded here or here.
We would like to express our gratitude for the codes, including Q-Align, which are the bases for building our project.
This project is for academic research purposes and released under the MIT license. Parts of this project use code, data, and models from other sources, which are subject to their respective licenses.