Repository for the paper Semantic Segmentation of Underwater Imagery: Dataset and Benchmark (IROS 2020)

• Paper (ArXiv) • Virtual Talk • SUIM Dataset • Experimental data • Checkpoints

SUIM Dataset

• For semantic segmentation of natural underwater images
• 1525 annotated images for training/validation and 110 samples for testing
• BW: Background/waterbody • HD: human divers • PF: Aquatic plants and sea-grass • WR: Wrecks/ruins
• RO: Robots/instruments • RI: Reefs/invertebrates • FV: Fish and vertebrates • SR: Sea-floor/rocks

SUIM-Net Model

• A fully-convolutional encoder-decoder network
• SUIM-Net (RSB): simple and light model; offers reasonable performance at a fast rate
• SUIM-Net (VGG): provides better generalization performance
• Detailed architecture is in models; associated train/test scripts are also provided
• The get_f1_iou.py script is used for performance evaluation

Benchmark Evaluation

• Performance analysis for semantic segmentation and saliency prediction
• SOTA models in comparison: • FCN • UNet • SegNet • PSPNet • DeepLab-v3
• Metrics: • region similarity (F score) and • contour accuracy (mIOU)
• Download experimental data from here and checkpoints data from here

Bibliography Entry

@inproceedings{islam2020suim,
  title={{Semantic Segmentation of Underwater Imagery: Dataset and Benchmark}},
  author={Islam, Md Jahidul and Edge, Chelsey and Xiao, Yuyang and Luo, Peigen and Mehtaz, 
              Muntaqim and Morse, Christopher and Enan, Sadman Sakib and Sattar, Junaed},
  booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year={2020},
  organization={IEEE/RSJ}
}

Acknowledgements

• https://github.com/qubvel/segmentation_models
• https://github.com/divamgupta/image-segmentation-keras
• https://github.com/floodsung/Deep-Learning-Papers-Reading-Roadmap
• https://github.com/zhixuhao/unet
• https://github.com/aurora95/Keras-FCN
• https://github.com/MLearing/Keras-Deeplab-v3-plus/
• https://github.com/wenguanwang/ASNet