Authors: Weihao Xia, Zhanglin Cheng, Yujiu Yang, Jing-Hao Xue.
Contact: [email protected]
We conduct experiments on different degradation types, which can be divided into two categories: one is the Regular Degradation Types, which contains the six common degradations including watermark, iirgular mask, noise, blur, JPEG compression and chromatic aberrations; the other is the Adverse Cityscapes under foggy, rainny and reflection conditions, respectively. With these two categories of degra- dations, we expect to validate the applicability of our method in both general scenes and a specific application–autonomous driving.
For Regular Degradation Types, please refer to data_generator for more details.
For Adverse Cityscapes, the fog and rain images can be downloaded from the official website of Cityscapes, which are available upon request.
We provide simulation scripts to generate window reflection. Reflection is a frequently-encountered source of image corruption that can arise when shooting through a glass surface. Below is a simulated sample using our provided scripts
We use pre-trained models to compute the input segmentation or parsing masks, i.e. DeepLabv3 model on Cityscapes dataset, ResNet50dilated model for Semantic Segmentation on MIT ADE20K dataset, DeepLabV2 on COCO-Stuff / PASCAL VOC 2012 dataset, and BiSeNet for face parsing on CelebAMask-HQ dataset.
Coming soon.
Coming soon.
Coming soon.
We conduct experiments on various datasets, which mostly can be found in our paper. Below are semantic segmentation and image restoration results of watermark and iirgular mask.
If you found our paper or code useful, please cite our paper.
@article{xia2019adverse,
author = {Weihao Xia and Zhanglin Cheng and Yujiu Yang and Jing-Hao Xue},
title = {Cooperative Semantic Segmentation and Image Restoration in Adverse Environmental Conditions},
year = {2019},
url = {http://arxiv.org/abs/1911.00679},
}