Voronoi-based VMAF for Omnidirectional Video

Requirements

Current implementation is based Python 3.

Dependencies:

Firstly, you should install the following dependencies:

• pip install wget
• pip install imageio
• pip install python-csv

Adding video files

Secondly, you should locate all distorted '.mp4' or '.yuv' files and a reference '.mp4' or '.yuv' file into the videos folder. warning Please make sure that all videos have the same resolution.

Testing

• python 360vmaf.py --w 3840 --h 2160 --f 100 --r sounders2

--w: resolution width of the videos, for example 3840

--h: resolution height of the videos, for example 2160

--f: number of frames, for example 100

--r: reference (original) .mp4 video name, for example sounders2.mp4, so you should not include the video type '.mp4' or '.yuv'

Results will be located in the main folder with the distorted video name and in .csv format. The script can generate two different quality scores: VMAF and VI-VMAF.

• VMAF is calculated based on equirectangular format.
• VI-VMAF is calculated based on the spherical voronoi technique which is explained in the paper: Voronoi-based Objective Quality Metrics for Omnidirectional Video. You can check the paper in doc folder.

Advanced options

• Changing VMAF model: In 360vmaf.py, you can change the VMAF model. The default model is vmaf_rb_v0.6.3. You can find other models in the folder of voronoiVMAF/model.

vmaf_model = 'vmaf_rb_v0.6.3/vmaf_rb_v0.6.3.pkl'

• Specifying number of Voronoi patches. By using option '--c', you can specify the number of patches, for example '--c 15'. The default parameter is 15, which was found an optimal number in the experiment. If you do not specify the number of patches, the script uses the default parameter.

python 360vmaf.py --w 3840 --h 2160 --f 100 --r sounders2 --c 15

Source codes for Voronoi diagram

We are providing source codes that contains the patch extraction. In the zip file there are two folders:

• PublishMaterial/: this folder contains MATLAB code (MATLAB/) that calls the exe file that extracts the patches, examples of XML configuration files that are read when extracting the patches (ConfigXMLExamples/), and pre-computed spherical Voronoi diagrams (SphericalVoronoiDiagrams) required for the extraction of the patches;
• OmniVideoQuality/: this folder contains the C++ code developed in MS Visual Studio for the extraction of the patches. It is the code for the generation of OmniVideoQuality.exe. See the Documentation/ subfolder on details about the configuration of the third-party libraries.