Deep Joint Image Filtering

MatConvNet implementation of our ECCV16 paper on joint image filtering.

Prerequisites

Tested on Linux or Windows
Matlab R2015b
MatConvNet

> matlab/vl_setupnn.m
> matlab/vl_compilenn.m

Training:

Generate the training data (nearly the same way used in SRCNN).

> examples/Train/generate_trainingdata.m

Or please download the training data (for 8x depth upsampling) and put it under the examples/Train/TrainingData/ folder.

> examples/Train/demo_train.m

Testing:

We provide our models for two tasks, i.e., depth map upsampling and Flash/Non-flash image noise reduction.

> examples/Test/cnn_test_upsampling.m
> examples/Test/cnn_test_noise_reduction.m

Note

For training with GPU, please uncomment line 215-217 and 222-223 in matlab/simplenn.m and then re-compile.

> vl_setupnn.m
> vl_compilenn('enableGpu', true)

Citation

@inproceedings{DJF-ECCV-2016,
    author = {Li, Yijun and Huang, Jia-Bin and Ahuja Narendra and Yang, Ming-Hsuan},
    title = {Deep Joint Image Filtering},
    booktitle = {European Conference on Computer Vision},
    year = {2016}
}

Acknowledgement

We express gratitudes to SRCNN as we benefit a lot from both their paper and codes.