CWG-RGB: Low-cost RGB Floating Garbage Dataset for Complex Urban Waterways

CWG-RGB is a low-cost RGB image dataset for floating garbage detection in complex urban waterways. The dataset contains 1,760 RGB images, including 1,176 target-containing images and 584 background negative images. A total of 3,557 floating-garbage instances are annotated in YOLO format across five categories: Bags-Films, Wood, Foam, Bottles-Cans and Dead Fish. Background negative images are provided with empty YOLO label files for false-positive suppression experiments and are not treated as an additional detection category. The repository also includes dataset split files, metadata, configuration files, benchmark scripts and reproduction notes.