Keras for DPNet

source code of DPNet paper

This project is writen by python2.7 and based on the Keras(2.1.4). The optical flow extractor is adopted from pyflow:

https://github.com/pathak22/pyflow

Running

Prepare the feature map

$ python feature_extraction.py 
         --test_video ./example/tr2.avi 
         --reference_video ./example/tr1.avi 
         --IQA SSIM 
         --patch_size 16 
         --save_file

Wait for the process finishing. Because of the optical flow operation is slow, it takes time...

The result will be stored in the "/tmp" folder.

Quality prediction

$ python video_test.py 
         --model ./model/LIVE_model_gmsd{32_8}.h5
         --feature_maps ./tmp/feaMap_tr2_GMSD.h5 
         --show_quality_changes

The overall quality score of the test video will be printed. The quality changes along with the time will also be provided, if the "--show_quality_changes" is chosen.