pnavarre / mgbpv2

MultiGrid BackProjection version 2.0 (MGBPv2) winner of perceptual track in ICCV AIM Extreme-SR Challenge 2019

Multi-Grid Back-Projection version 2.0 (MGBPv2)

Citation

Pablo Navarrete Michelini, Wenbin Chen, Hanwen Liu and Dan Zhu, "MGBPv2: Scaling Up Multi-Grid Back-Projection Networks", in The IEEE International Conference on Computer Vision Workshops (ICCVW), 2019.

BibTeX

@inproceedings{MGBPv2,
    title     = {{MGBP}v2: Scaling Up Multi--Grid Back--Projection Networks},
    author    = {Navarrete~Michelini, Pablo and Chen, Wenbin and Liu, Hanwen and Zhu, Dan},
    booktitle = {The {IEEE} International Conference on Computer Vision Workshops ({ICCVW})},
    month     = {October},
    year      = {2019},
    url       = {https://arxiv.org/abs/1909.12983}
}

AIM-2019 Output Images (DIV8K Test Set)

• 16✕ Test Set BOE-Perceptual (1^st place) (11GB)
• 16✕ Test Set BOE-Fidelity (5^th place) (11GB)

AIM-2019 Model Files

• BOE-Perceptual (1^st place) (112.7MB)
• BOE-Fidelity (5^th place) (1.1GB)

Instructions:

• Download model files using ./download_models.sh or links above.

• Copy input images in input (provided as empty directory).

• To upscale images 16x run: python run.py.

  Output images will come out in output (automatically created and cleaned if already exists).

• The GPU number, model file and memory target can be changed in run.py (in comment "CHANGE HERE").

Requirements:

• Python 3, PyTorch, NumPy, Pillow, OpenCV