CaUS_Visibility_Artifacts

Repository for the paper "Neural Networks for Classification and Unsupervised Segmentation of Visibility Artifacts on Monocular Camera Image"

Dataset

Dataset "VisibilityArtifacts" consists of a partial combination of images from the data sets:

1. Woodscape ("Soiling Detection" sample), radial distortion has been eliminated for this set and the images have been corrected;
2. DrivingStereo ("Different weathers" sample);
3. TapmerDetection;
4. ACDC.

In order to balance the number of images per class, data augmentation was performed using the imgaug tool, from thoseimages in which there were no visibility artifacts. As a result, a balanced "VisibilityArtifacts" dataset was formed, including 22311 images, divided into 7 categories, it's description is given in Table 2, examples of images are shown in Fig. 2 in the article.

If you want to use this dataset in your research, then read the following conditions:

1. Woodscape license;
2. DrivingStereo license;
3. ACDC license.

Because if you download the dataset "VisibilityArtifacts", then you accept them automatically.

Links to the dataset:

1. Classification: part 1, part 2;
2. Segmentation: part 3.

For comfortable data processing, use the code from the "Loading data and splitting it" section of the notebook "Training Classifiers.ipynb".

Citation

@article{kuznetsov2022neural,
  title={Neural Networks for Classification and Unsupervised Segmentation of Visibility Artifacts on Monocular Camera Image},
  author={Kuznetsov, Vladislav I and Yudin, Dmitry A},
  journal={Optical Memory and Neural Networks},
  volume={31},
  number={3},
  pages={245--255},
  year={2022},
  publisher={Springer}
}