pnavarre/pirm-sr-2018

Generative Multigrid Backprojection Super Resolution (G-MGBP) 2nd best perceptual quality in PIRM SR Challenge

Multigrid Backprojection Super Resolution

Pablo Navarrete Michelini, Dan Zhu and Hanwen Liu

Citation

Pablo Navarrete Michelini, Hanwen Liu and Dan Zhu, "Multigrid Backprojection Super-Resolution and Deep Filter Visualization", to appear in the Thirty-Third AAAI Conference on Artificial Intelligence (AAAI-19)

Pablo Navarrete Michelini, Dan Zhu and Hanwen Liu, "Multi-Scale Recursive and Perception-Distortion Controllable Image Super-Resolution", in The European Conference on Computer Vision Workshops (ECCVW), 2018.

BibTeX MGBP (AAAI-19)

@inproceedings{MGBP,
    title        = {Multigrid Backprojection Super--Resolution and Deep Filter Visualization},
    author       = {Navarrete~Michelini, Pablo and Liu, Hanwen and Zhu, Dan},
    booktitle    = {Proceedings of the Thirty--Third AAAI Conference on Artificial Intelligence (AAAI 2019)},
    year         = {2019},
    organization = {AAAI},
    url          = {http://arxiv.org/abs/1809.09326}
}

BibTeX Generative MGBP (ECCVW)

@inproceedings{G-MGBP,
    author    = {Navarrete~Michelini, Pablo and Liu, Hanwen and Zhu, Dan},
    title     = {Multi--Scale Recursive and Perception--Distortion Controllable Image Super--Resolution},
    booktitle = {The European Conference on Computer Vision Workshops (ECCVW)},
    month     = {September},
    year      = {2018},
    url       = {http://arxiv.org/abs/1809.10711}
}

PIRM-2018 Output Images

• 4✕ Test set BOE-R3 (2^nd place) (63.2MB)
• 4✕ Test set BOE-R2 (5^th place) (48.8MB)
• 4✕ Test set BOE-R1 (7^th place) (45.3MB)

Instructions:

• Copy low resolution images from the test set in input_images (provided as empty directory)
• Run python main.py
• Upscale images will come out in output_images (automatically created and cleaned if already exists)
• By default BOE-R3 model is used. The GPU number and model parameters can be changed in main.py.

Requirements:

• Python 3, PyTorch, NumPy, Pillow