CANSURF: An ASV-View Can Dataset and Benchmark for Detection and Tracking of Surface-Level Debris

Dataset which contains the raw CANSURF images and labels. Augmentation script and the latest version of the dataset can be found in the following Github repository: https://github.com/ZaidAljundiHW2/CANSURF CANSURF-v2 reorganises the split in a per-video fashion where video frames of each video are placed together in the training and validation subdirectories instead of having them spread out through both. The ratio between training and validation data was also adjusted to 70/30. CANSURF-v2 also features a higher number of original content from the authors to replace a portion of the open-source Roboflow dataset content present in CANSURF-v1. Benchmarking with YOLOv11 showed improved precision and recall. The total image count has increased to 7,171.