vpx-custom-enc

run exp

• Pre-requisites
  • ffmpeg (if you are using ffmpeg binary/docker, make sure your ffmpeg is aliased to ffmpeg and callable in bash)
  • python 3.x with numpy, math
• Data placement

~/dataset-all
      |--------sr #should contain quality.json
      |--------hr

• Run test with one click : test on (10,20,30M) bitrates using ffmpeg/vpxenc

cd script
bash run_all.sh ~/dataset-all

• Result placements

~/dataset-all
      |--------sr_yuv
      |--------sr_{bitrate}_vpx.webm
      |--------sr_{bitrate}_vpx.log #logs fps
      |--------sr_{bitrate}_vpx_quality.log #logs quality
      |--------sr_{bitrate}_vpx #contains pngs extracted from vpx encoded video

      |--------sr_{bitrate}_ffmpeg.webm
      |--------sr_{bitrate}_ffmpeg_quality.log #logs quality
      |--------sr_{bitrate}_ffmpeg #contains pngs extracted from ffmpeg video

• Result example
• Experiment result summary @ Intel Core i9-9900K, 3.6 GHz, 4 core, 1 CPU (iris1.kaist.ac.kr)

FPS	Quality

• Experiment result summary @ AWS g4dn.metal

FPS	Quality

VP8 FPS	VP9 FPS

testing videos

cd script
bash ffmpeg_test.sh [content] #make using ffmpeg  
bash vpx_test.sh [content] #make using vpx   
bash compare_psnr.sh [content] #get psnr results  

results

• Hearthstone (1080p, 60fps)

  • my_vpxenc
    PSNR y:39.433927 u:42.226109 v:43.415964 average:40.290927 min:32.684165 max:47.098691
    200fps
  • ffmpeg
    PSNR y:39.494910 u:42.304841 v:43.497325 average:40.355452 min:32.424708 max:50.370501

• CSGO (1080p, 60fps)

  • my_vpxenc
    PSNR y:32.192989 u:40.650755 v:42.266148 average:33.699856 min:30.047399 max:55.728286
    121fps
  • ffmpeg
    PSNR y:32.321720 u:40.661769 v:42.278743 average:33.821850 min:30.023484 max:55.747964