Bokeh effect 学习与研究

基于SBTNet进行了研究学习，结合DPT（Midas 3.0）对模型进行了改进。

SBTNet源码地址，下载pretrained model，请放到checkpoints目录下

https://github.com/JuewenPeng/SBTNet

Intel lab DPT源码地址，下载pretrained model，请放到model_cache目录下

https://github.com/intel-isl/DPT

运行SBTNet模型的命令

python evaluation.py --root_folder TEST_ROOT_FOLDER --save_folder SAVE_FOLDER

运行改进模型的命令，光圈根据实际修改调整

python demo.py --src_f 16.0 --tgt_f 1.8

以下为SBTNet的原始README.MD

*************************************************************************************

SBTNet: Selective Bokeh Effect Transformation

Our solution in competition NTIRE 2023 Bokeh Effect Transformation: https://codalab.lisn.upsaclay.fr/competitions/10229.

Test Results

Download the test results from Google Drive.

Usage

Download the pretrained model from Google Drive, and place it in the folder checkpoints. Run the following code to generate test results.

python evaluation.py --root_folder 'TEST_ROOT_FOLDER' --save_folder 'SAVE_FOLDER'

• root_folder: root folder of the test dataset.
• save_folder: folder to save the results.

Citation

If you find our work useful in your research, please cite our paper.

@inproceedings{Peng2023Selective,
  title = {Selective Bokeh Effect Transformation},
  author = {Peng, Juewen and Pan, Zhiyu and Liu, Chengxin and Luo, Xianrui and Sun, Huiqiang and Shen, Liao and Xian, Ke and Cao, Zhiguo},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
  year = {2023}
}