AquaSurf-Malnad-223: A vegetation-stratified benchmark for false-positive analysis in freshwater debris detection

The Dataset has 223 geotagged photographs of vegetation-covered ponds, village tanks and slow streams around Shivamogga, Karnataka, India. Every image carries a human rating of floating-vegetation coverage (none, low, medium, high). The release contains 84 verified debris boxes (75 plastic, 7 paper, 2 glass) in 26 images, 197 images verified to contain no debris, the 817 raw proposals of a COCO-pretrained Faster R-CNN used for the reference evaluation, a per-image manifest (GPS, capture time, vegetation rating, enhancement metrics), CSV and COCO annotations, a datasheet and SHA-256 checksums. Bystanders in 21 images are irreversibly redacted; the data must not be used for person detection. Recall is not verified, so the data should not be used to estimate debris abundance. The images and annotations are identical to IEEE DataPort record 10.21227/k8d4-5n64; this version fills in the DOI placeholder left in CITATION.cff and DATASHEET.md.