Fast Image Processing with Fully-Convolutional Networks
This is a Tensorflow implementation of Fast Image Processing with Fully-Convolutional Networks.
Required python libraries: Tensorflow (>=1.0) + Opencv + Numpy.
Tested in Ubuntu + Intel i7 CPU + Nvidia Titan X (Pascal) with Cuda (>=8.0) and CuDNN (>=5.0). CPU mode should also work with minor changes.
Quick Start (Testing)
- Clone this repository.
- Run "CAN24_AN/demo.py". This will generate results on L0 smoothing in "CAN24_AN/L0_smoothing/MIT-Adobe_test_1080p_result".
- To test a different model, change the variable "task" in "demo.py"
- To train, change "is_training" to "True".
- To set up a customized training procedure, change the file paths in "prepare_data()". See the commands in the code.
- The single network for all operators is "combined.py" in the folder "Single_Network". Run it and its result is in "Single_Network/result_combined/video".
- The parameterized network is "parameterized.py" in the folder "Parameterized_Network". Run it and its result is in "Parameterized/result_parameterized/video".
If you want to experiment on the data in our evaluation, please email to email@example.com.
If you use our code for research, please cite our paper:
Qifeng Chen, Jia Xu, and Vladlen Koltun. Fast Image Processing with Fully-Convolutional Networks. In ICCV 2017.