ShenZheng2000 / DarkCityScape_mIOU_mPA

DarkCityScape_mIOU_mPA

This is a repository for calculating the mIOU (mean intersection over union) and mPA (mean pixel accuracy) for DarkCityScapes. DarkCityScapes is the "Dark version" of the CityScape dataset [1] and is specifically designed for extreme low-light image enhancement tasks. The dataset is simulated using gamma correction on 150 CityScape validation images.

Sample Images from DarkCityScapes

Get Started

1. Download the DarkCityScape dataset from Baiduyun with passord wvhy

2. Download the DarkCityScape labels from Baiduyun with password a2z5

3. Put the segmentation outcome of the enhanced images in path_to_your_pred And the Groudtruth segmentation labels in path_to_your_gt

4. Run the following script

python main.py --pred path_to_your_pred --gt path_to_your_gt

Note:

• You could refers to Dark.sh if you are not sure how to put the images.
• The size of pred and gt should be same.
• The result of mIOU and mPA will be in output_iou.txt and output_pa.txt, respectively.

Sample Result

Following is the result table from some state-of-the-art low-light image enhancement models

	Dark	PIE [2]	Retinex [3]	MBLLEN [4]	KinD [5]	ZeroDCE [6]
mIOU	54.49%	61.97%	57.96%	51.98%	63.42%	64.36%
mPA	70.76%	68.89%	66.76%	59.06%	71.69%	74.20%

References:

[1] Cordts, Marius, et al. "The cityscapes dataset for semantic urban scene understanding." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.

[2] Fu, Xueyang, et al. "A probabilistic method for image enhancement with simultaneous illumination and reflectance estimation." IEEE Transactions on Image Processing 24.12 (2015): 4965-4977.

[3] Wei, Chen, et al. "Deep retinex decomposition for low-light enhancement." arXiv preprint arXiv:1808.04560 (2018).

[4] Lv, Feifan, et al. "MBLLEN: Low-Light Image/Video Enhancement Using CNNs." BMVC. 2018.

[5] Zhang, Yonghua, Jiawan Zhang, and Xiaojie Guo. "Kindling the darkness: A practical low-light image enhancer." Proceedings of the 27th ACM international conference on multimedia. 2019.

[6] Guo, Chunle, et al. "Zero-reference deep curve estimation for low-light image enhancement." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020.