MoLa-VI Dataset
Description
This repository presents the dataset described in the article "In-car Damage Dirt and Stain Estimation with RGB Images", published in Proceedings of the 13th International Conference on Agents and Artificial Intelligence (2021). The dataset consists of two main folders (IMAGES and SEGMENTATIONS) and a .json file (dataset.json). In the IMAGES folder, we find 135 cars duly separated by 135 folders, in each one of them we find images of the interior of the respective cars under 9 different views duly divided by folders (P1 to P9). In the SEGMENTATIONS folder, we find the same structure with the car image masks found in the IMAGES folder, these masks are composed of the classes found in the IMAGES, identified by a per-pixel id: (1) DMG_CUT, (2) DMG_WEAR, (3) DMG_BROKEN, (4) STAIN, (5) DIRT, and (6) GOOD. It is also available a .json file with all the information about the dataset, that is, with an ID associated with each car, in which for each one of them presents the information of the car model (i.e. brand, type, seat colour, plastics colour, ceiling colour), and with information regarding the image and segmentation, in "*.jpeg" format. At the beginning of the .json file, the dataset location path must be updated, to index all the dataset information. In the dataset folder there are also 4 .mat files: (1-FixRoot) change the dataset root directory; (2-FilterDataset_percent) create 3 sub-json files with random samples and fixed percentage quantities; (3-Json2OBD) exports dataset samples, from a json file, with a sample format compatible with object detectors; (3-Json2SEG) exports dataset samples, from a json file, with a sample format compatible with segmentators. For further information see https://repositorium.sdum.uminho.pt/handle/1822/74435.