BRA-Dataset is a dataset about the animals that most suffer accidents on Brazilian highways. It has 5 classes of medium and large animals, with 1823 imagens. It is free and open-source.
Data Augmentation script with geometric transformations
Tutorial for COCO format conversion
1823 Images Total
2060 Labels Total
Classes:
Tapir----------------------------------------------> 403 Images / 478 Labels
Jaguarundi-----------------------------------------> 311 Images / 341 Labels
Maned Wolf-----------------------------------------> 340 Images / 410 Labels
Puma-----------------------------------------------> 476 Images / 495 Labels
Giant Anteater-------------------------------------> 293 Images / 336 Labels
clone the repository YOLO2COCO
With the diretory struture
Dataset
---images
---labels
---class.names
---gen_config.data
---train.txt
---valid.txt
execute the command in YOLO2COCO folder: python darknet2coco.py --data_path ../Dataset/gen_config.data
https://mega.nz/file/3xREFRrJ#pAmxuOanRehFlo75LUAfBokq2huZ22jUfcuPRFNQ5ow - For Tapir Class with oclusion problem
https://mega.nz/file/PhhFhSgQ#1NScqSqNIOcpWgE7JR_ntbogw-b7ecKdle0mn6PRSjI - For Giant Anteater with small object and camouflage
This project is a master's project in computer science and computational mathematics at the University of São Paulo - Brazil (PPG-CCMC, ICMC-USP)