DDDF-Bokeh-Effect-Rendering-ICCV2019

Code for the paper "Depth-guided dense dynamic filtering network for bokeh effect rendering", ICCV Workshop 2019

Runner-Up Award Winner in AIM 2019 Bokeh Effect Challenge

Certificate: https://data.vision.ee.ethz.ch/cvl/aim19/AIM2019_award_certificates.pdf

Challenge Report: IEEE Paper Link , PDF

PreTrained Model

Google Drive Link

Usage

1. Place the input test images in the folder: Set14/LR

2. Change the directory to 'MegaDepth' and run this command (requires pytorch):

python demo_padding.py 

3. Change the directory to 'Salient_Object_Detection' and run this command (requires tensorflow):

python inference.py --rgb_folder=../Set14/LR

4. Change the directory to 'src' and run the final inference command:

python main.py --data_test MyImage --model sm_space2depth_densedecoder_instancenorm_seg_depth_beginning_dynamic_filter_separatedecoder --scale 1 --pre_train ./trained_model/model_latest.pt --test_only --save_results --save 'upload' --testpath ../ --testset Set14

This will generate the final results in the folder: SR/BI/upload/results

Credits

1. Depth Estimation module adopted from here
2. Saliency Detection module adopted from here

Citation

If you find our paper/results helpful in your research or work please cite our paper.

@inproceedings{purohit2019depth,
  title={Depth-guided dense dynamic filtering network for bokeh effect rendering},
  author={Purohit, Kuldeep and Suin, Maitreya and Kandula, Praveen and Ambasamudram, Rajagopalan},
  booktitle={2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)},
  pages={3417--3426},
  year={2019},
  organization={IEEE}
}