Pablo Navarrete Michelini, Dan Zhu and Hanwen Liu
@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}
}
@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}
}
- 4✕ Test set BOE-R3 (2nd place) (63.2MB)
- 4✕ Test set BOE-R2 (5th place) (48.8MB)
- 4✕ Test set BOE-R1 (7th place) (45.3MB)
- 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.
- Python 3, PyTorch, NumPy, Pillow