Deep_SESR

Deep_SESR copied to clipboard

SESR: Simultaneous Enhancement and Super-Resolution

• Perceptually enhanced image generation at higher spatial scales

Pointers:

• Paper: http://www.roboticsproceedings.org/rss16/p018.pdf
• Preprint: https://arxiv.org/pdf/2002.01155.pdf
• RSS-2020 Spotlight Talk: https://youtu.be/8zBdFxaK4Os
• Data: http://irvlab.cs.umn.edu/resources/ufo-120-dataset

Deep SESR model

• An efficient model for underwater imagery; can be trained end-to-end for 2x-4x SESR
• Model architecture and implementation details: https://arxiv.org/pdf/2002.01155.pdf
• Weights for Deep SESR 2x with 1D FENet (trained on UFO-120) are provided in models/
  • HDF5: deep_sesr_2x_1d.h5 file; use test_sesr_Keras.py
  • Protobuf: deep_sesr_2x_1d.pb file; use test_sesr_TF.py

UFO-120 dataset

• 1500 training and 120 test samples (underwater images)
• Facilitates paired training of 2x, 3x, and 4x SESR models
• Also has annotated saliency maps for training saliency prediction models
• Can be downloaded from: http://irvlab.cs.umn.edu/resources/ufo-120-dataset

Bibliography entry:

@inproceedings{islam2020sesr,
    author={Islam, Md Jahidul and Luo, Peigen and Sattar, Junaed},
    title={{Simultaneous Enhancement and Super-Resolution of Underwater Imagery 
    	    for Improved Visual Perception}},
    booktitle={Robotics: Science and Systems (RSS)},
    year={2020},
    month={July},
    address={Corvalis, Oregon, USA},
    doi={{10.15607/RSS.2020.XVI.018}}
}