This is one of the approaches participating in AIM 2019 challenge on bokeh effect synthesis. For more methods and results, please refer to the summary.
Save the pre-trained models to checkpoints/bokeh_saliency/
You can directly download them here.
Or you can follow the open-source saliency detection methods to obtain the masks on your own.
Then you should save them to [path to masks]
python generate.py --name bokeh_saliency --which_epoch 350 --dataset_mode custom --origin_dir [path to original images] --masks_dir [path to masks]
If you find this project useful in your research, please consider citing:
@INPROCEEDINGS{9022578,
author={A. {Ignatov} and J. {Patel} and R. {Timofte} and B. {Zheng} and X. {Ye} and L. {Huang} and X. {Tian} and S. {Dutta} and K. {Purohit} and P. {Kandula} and M. {Suin} and A. N. {Rajagopalan} and Z. {Xiong} and J. {Huang} and G. {Dong} and M. {Yao} and D. {Liu} and M. {Hong} and W. {Lin} and Y. {Qu} and J. {Choi} and W. {Park} and M. {Kim} and R. {Liu} and X. {Mao} and C. {Yang} and Q. {Yan} and W. {Sun} and J. {Fang} and M. {Shang} and F. {Gao} and S. {Ghosh} and P. K. {Sharma} and A. {Sur} and W. {Yang}},
booktitle={2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)},
title={AIM 2019 Challenge on Bokeh Effect Synthesis: Methods and Results},
year={2019},
volume={},
number={},
pages={3591-3598},
doi={10.1109/ICCVW.2019.00444}}