Dataset
Underwater Seafloor Litter Drone Image Database
Labeled drone images of seafloor litter for training object-detection models. Available on request for research and computer vision work.
Original RAW
The original images, captured with a DJI Mavic 2 Pro (Hasselblad camera, 20 MP, 4K UHD) and a DJI Air 2S (2.4 µm pixel pitch, 5.4K UHD).
| Key | Value |
|---|---|
| Number of images | 5 405 |
| Number of labels | 12 786 |
| Size | ~193 GB |
| Format | DNG |
| Resolution | 5568 x 3648, 5568 x 3078, 5472 x 3648 |
| Classes | 31 (Fragments, Construction material, Tires, Cans, Bottles, Ropes, Cloth, Bags, Fishing tools, Caps & lids, Cups, Buckets, Pipe, Footwear, Cartons, Dishes, Paper, Chain, Chair, Shopping carts, Trash can, Road sign, Dumpster, Boats, Glove, Car parts, Traffic cone, Flipper, Scooter, Electric drill, Cigarette butt) |
| Annotations | COCO, Voxel51 |
| Preview | No preview available |
Half Size
The same images halved in width and height and converted to PNG for faster handling. Every other subset is derived from this one.
| Key | Value |
|---|---|
| Number of images | 5 405 |
| Number of labels | 12 786 |
| Size | ~33 GB |
| Format | PNG |
| Resolution | 2732 x 1820, 2732 x 1535 |
| Classes | 31 (Fragments, Construction material, Tires, Cans, Bottles, Ropes, Cloth, Bags, Fishing tools, Caps & lids, Cups, Buckets, Pipe, Footwear, Cartons, Dishes, Paper, Chain, Chair, Shopping carts, Trash can, Road sign, Dumpster, Boats, Glove, Car parts, Traffic cone, Flipper, Scooter, Electric drill, Cigarette butt) |
| Annotations | COCO, Voxel51 |
| Preview | Note: Preview is in JPG format to speed up loading |
All Patches
Each annotation is cropped to a 640×640 patch; larger annotations are scaled to fit. Nearby objects share a patch, so the same object can appear in more than one image. Duplication is kept to a minimum: an object is only added if it is not already covered. Labels at the edge of a patch may be cropped.
| Key | Value |
|---|---|
| Number of images | 6 298 |
| Number of labels | 17 913 |
| Size | 3.3 GB |
| Format | PNG |
| Resolution | 640 x 640 |
| Classes | 31 (Fragments, Construction material, Tires, Cans, Bottles, Ropes, Cloth, Bags, Fishing tools, Caps & lids, Cups, Buckets, Pipe, Footwear, Cartons, Dishes, Paper, Chain, Chair, Shopping carts, Trash can, Road sign, Dumpster, Boats, Glove, Car parts, Traffic cone, Flipper, Scooter, Electric drill, Cigarette butt) |
| Annotations | COCO, Voxel51 |
| Preview | Note: Preview is in JPG format to speed up loading |
Selected Patches
A curated subset limited to the nine classes with enough examples to be represented properly.
| Key | Value |
|---|---|
| Number of images | 4 016 |
| Number of labels | 10 958 |
| Size | 2.1 GB |
| Format | PNG |
| Resolution | 640 x 640 |
| Classes | 9 (Fragments, Construction material, Tires, Cans, Bottles, Cloth, Bags, Caps & lids, Cups) |
| Annotations | COCO, YOLO YAML (with train/val/test split) |
| Preview | Note: Preview is in JPG format to speed up loading |
Access
The full set is close to 500 GB, so we do not keep a public download online. Write to with the subset you need (original RAW, half size, or patches) and a short note on how you plan to use it, and we will send you a transfer link.
