Due to highly variying domain features of different underwater enviornment, the publically available datasets alone are not the best fit to train a deep learning algorithm to predict trash. Therefore we propose a cumulatuve, self-annonated dataset that provides a good foundation for training models to detect and classify trash underwater, and also provide benchmarks for the same. This repository aims to generate code that detects trash, classifies it into plastic / trash / underwater debris against other factors like fish / flora & fauna and input rover images.
Aiming at the problem of insufficient storage space and limited computing ability of underwater mobile devices, an underwater garbage detection algorithm is proposed.
since YoloV8 is quite new and still under development it had a lot of bugs and didnt give the best outputs.
Gave more accurate detections from v5 and lower false positives but better accuracy at the cost of lower detections.
Higer detections, higher false negatives relative to v6 but more detections.