MoLa-VI Dataset

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.

本存储库展示了发表于《第十三届国际智能体与人工智能会议》（2021年）的论文《基于 RGB 图像的车内损伤、污渍估计》所描述的数据集。该数据集包含两个主要文件夹（IMAGES 和 SEGMENTATIONS）以及一个 . 文件（dataset.）。在 IMAGES 文件夹中，我们发现了135辆汽车，每辆汽车均单独放置于135个文件夹中，每个文件夹内包含从9个不同视角（P1至P9）拍摄的对应汽车内部图像。在 SEGMENTATIONS 文件夹中，结构与此相同，包含与 IMAGES 文件夹中的汽车图像相同的掩码，这些掩码由在 IMAGES 中发现的类别组成，并按每像素的 ID 进行标识：（1）DMG_CUT，（2）DMG_WEAR，（3）DMG_BROKEN，（4）STAIN，（5）DIRT 和（6）GOOD。此外，还提供了一份包含有关数据集全部信息的 . 文件，其中每个汽车都与一个 ID 相关联，其中详细列出了每辆汽车的信息，包括车型（即品牌、类型、座椅颜色、塑料颜色、天花板颜色），以及有关图像和分割的信息，格式为“*.jpeg”。在 . 文件的起始部分，需更新数据集位置路径，以索引所有数据集信息。在数据集文件夹中还包括4个 .mat 文件：（1-FixRoot）更改数据集根目录；（2-FilterDataset_percent）创建3个子 . 文件，包含随机样本和固定百分比的样本数量；（3-Json2OBD）从 文件导出与对象检测器兼容的样本格式；（4-Json2SEG）从 文件导出与分割器兼容的样本格式。有关更多信息，请参阅 https://repositorium.sdum.uminho.pt/handle/1822/74435。