WaveOne I-frame Compression Demo

Note: Better viewed on a large screen

We asked subjectify.us to perform a user study of the WaveOne codec for a total of 6600 comparisons. We benchmarked against slow but powerful reference implementations of HEVC, VVC and AV1. The result of the study is as follows:

At the same bitrate, WaveOne reconstructions are 9.2x more likely to be chosen over HEVC, 9.8x more likely to be chosen over AV1 and 4.6x more likely to be chosen over VVC.
Even when consuming half the bitrate, WaveOne reconstructions are is 2.0x more likely to be chosen over HEVC, 1.9x more likely to be chosen over AV1, and 1.3x to be chosen as VVC.

Full details of how we set up our study and other details can be found below.

Best viewed on a big screen.

On iPhone, to move the slider while zoomed in, tap and hold until you get haptic feedback!

Kilobyte (kB)

Bits per pixel (bpp)

Study Details

The test dataset included the first 50 images from the CLIC image dataset. The images were center-cropped to 512x512, as required by the study to better study the effect of distortions. Each image pair was rated by approximately 20 people in an "2AFC (two alternatives, forced choice)" experiment with the baseline codec randomly placed, for a total of 6600 comparisons.

We benchmarked against HEVC (HM v16.23), VVC (VTM v1.1.0) and AV1 (SVT-AV1 v1.1.0) which are the very slow but powerful reference implementations. We exclude the container headers (mp4), but include the codec headers when computing the file size of the single-frame video for both WaveOne and the baselines.

The percentage of times WaveOne reconstructions were chosen (vs baseline codecs) is presented below. The error-bars correspond to 95% confidence intervals.