Service: VMAF Studio - EyevinnOSC/community GitHub Wiki

VMAF Studio

VMAF Studio (serviceId ablindberg-osc-vmaf-studio) is a web-based interface for running VMAF (Video Multi-Method Assessment Fusion) video quality analysis on OSC. It uses the EasyVMAF service to compare a reference video against a transcoded version and compute a perceptual quality score. In Eyevinn Open Source Cloud (OSC) it is a long-running service instance. License: MIT. Source: ABLindberg/osc-vmaf-studio.

VMAF is Netflix's open source quality metric, widely used to validate encoded renditions before publishing. A score of 100 means the distorted video is perceptually identical to the source.

What it can do

  • Compare a reference video with a transcoded rendition by URL and compute the VMAF score.
  • Show the aggregate VMAF score and a per-frame score chart.
  • Work with video files reachable by URL or stored in a MinIO bucket on OSC.

Creating an instance

You need your OSC personal access token (create it on the API Access page, see Finding Your API Token); VMAF Studio uses it to call EasyVMAF on your behalf. Store it as a service secret with create-service-secret (serviceId: "ablindberg-osc-vmaf-studio", name osctoken). Then call create-service-instance with serviceId: "ablindberg-osc-vmaf-studio". The instance name must be alphanumeric and at most 20 characters.

{ "name": "myvmaf", "oscAccessToken": "{{secrets.osctoken}}" }

Via CLI:

npx -y @osaas/cli secret set vmafstudio osctoken <your-personal-access-token>

npx -y @osaas/cli create ablindberg-osc-vmaf-studio myvmaf \
  -o oscAccessToken="{{secrets.osctoken}}"

Wait with wait-for-service-ready (or get-instance-status) and read the URL with describe-service-instance.

Options

Option Required Type Meaning
name Yes string Name of the instance. Alphanumeric/underscore only, at most 20 characters.
oscAccessToken No string, sensitive OSC personal access token used to call the EasyVMAF service that performs the analysis. Used server-side only and never sent to the browser. The VMAF Studio UI also lets you set it in its Settings panel as "OSC Token". Pass as {{secrets.NAME}}.

A {{secrets.NAME}} reference must be the entire field value. Overwriting a secret does not restart a running instance.

Using the instance

Open the VMAF Studio URL and:

  1. Enter the URL of a reference video (the original, uncompressed or high-quality source).
  2. Enter the URL of a distorted video (the transcoded rendition to evaluate).
  3. Click Analyze.

VMAF Studio submits the job to EasyVMAF, polls for completion and shows the aggregate VMAF score and the per-frame chart.

Interpreting results

Score range Interpretation
95-100 Excellent, visually transparent
85-94 Good, suitable for most streaming delivery
70-84 Acceptable, noticeable quality loss on large screens
< 70 Poor, visible artefacts, consider re-encoding

Resources