Dataset for semantic segmentation of underwater imagery.
- Paper: https://arxiv.org/pdf/2004.01241.pdf
- Code: https://github.com/xahidbuffon/SUIM-Net
- Project page: https://irvlab.cs.umn.edu/image-segmentation/suim
| Object category | Symbol | RGB color code |
| Background (waterbody) | BW | 000 (black) |
| Human divers | HD | 001 (blue) |
| Aquatic plants and sea-grass | PF | 010 (green) |
| Wrecks and ruins | WR | 011 (sky) |
| Robots (AUVs/ROVs/instruments) | RO | 100 (red) |
| Reefs and invertebrates | RI | 101 (pink) |
| Fish and vertebrates | FV | 110 (yellow) |
| Sea-floor and rocks | SR | 111 (white) |
Folder structure:
- train_val/ contains 1525 paired samples for training/validation
- images/: RGB images of underwater scenes
- masks/: segmentation labels
- Each RGB color represents a different object category
- Further details in the paper (Section III)
- TEST/ contains 110 paired samples for benchmark evaluation
- images/: RGB test images
- masks/: ground truths labels
- Combined RGB masks are provided
- Individual binary masks are also provided in separate folders
- Benchmark_Evaluation/ contains all the experimental data for the SOTA models