videocodecs.info benchmark

This repo contains the source code for benchmarks and frontend of the videocodecs.info website. Under vivict/ directory, there's a custom version of vivict created by Gustav Grusell. Library used to draw graphs is highcharts.js.

This codebase is a subject to change. Right now that's just a couple of scripts glued together. The project should use better benchmark runner in the future. Preferably with Ansible so that the user doesn't need to install everything manually.

Install requirements

amd64 Linux Host (preferably Arch)

Dependencies:

• python
• python-psutil
• python-numpy
• python-dataclasses-json
• ffmpeg
• vmaf
• ssimulacra2_rs
• npm
• docker
• any other program specified in benchmark config (vvenc, uavs3e, ...)

How to use

Modify benchmark config

Benchmark configuration is imported in the run_benchmark.py file. By default it's set to benchmark_config_fhd.py. You might also want to review config_common.py.

Set input parameters

Modify the config_input.py. Here's an example of how it should look like

input_clip = {
  'clipname': 'matrix',
  'title': 'Matrix',
  'desc': 'some scenes from Matrix movie trailer',
  'src': 'https://archive.org/download/the-matrix-resurrections-official-trailer-prores/rev-1-dom_trailer_1_online_txtd_mtrx4_ProResMaster_no_slate_prores.mov',
  'gop': 128,
  'fps': 23.98,
  'frames': 360,
  'width': 1920,
  'height': 1080,
}

work_dir = '/tmp'  # directory where the input .y4m file should be
final_dir = '/mnt/codec_benchmarks'  # directory for processed clips
cpu_cost_global_mult = 1.0  # adjust for more/less CPU demanding encodes

Run benchmark

Place input .y4m file in the work_dir.

Run normally

$ python run_benchmark.py

Only produce stats (vmaf, ssimulacra2, frame bitrate, ...) this assumes all clips are already encoded

$ python run_benchmark.py --skip-encoding

Only process collections which aren't present in final_dir

$ python run_benchmark.py --update

Build Vivict

$ cd vivict/
$ npm install
$ npm run build
$ cd ..

Start the frontend

$ docker build -t vci-nginx-standalone .
$ docker run -p 80:80 -v /mnt/codec_benchmarks:/www/data:ro vci-nginx-standalone

It should be now accessible under http://localhost