Mezzframe AI · Open source

An open source lab for deep video engineering.

Mezzframe AI uses AI to study and improve how moving pictures are captured, encoded, carried across networks, understood and kept for the long run.

Get involved See the code
Headlines
  • Now in the lab: Encode, Live, Preserve, Meaning
  • Capture joins the lineup later
  • Code Apache-2.0 · Writing CC BY 4.0
  • If you can't rerun it, we haven't finished it
  • AI proposes, people approve
  • Mezzframe AI is looking for video engineers, AI builders and archivists
Guide

What we work on

  1. Encode

    Compression, codecs and color. Rate control, quality metrics, learned codecs, and AI that helps a classic encoder make better choices.

    • codecs
    • rate control
    • VMAF
    • learned compression

  2. Live

    Video over IP for broadcast and streaming: ST 2110, NDI, PTP, NMOS, SRT and RIST. Anything that touches a live system asks a person first.

    • ST 2110
    • PTP
    • NMOS
    • SRT
    • RIST

  3. Preserve

    Legacy video and motion picture film. AI that helps analyze, restore and archive older formats while the original stays the original.

    • restoration
    • film scans
    • archival formats

  4. Meaning

    Metadata, captions, provenance and video understanding. Knowing what is in a frame and where it came from.

    • captions
    • provenance
    • C2PA
    • scene understanding

  5. CaptureComing later

    Cameras, sensors and image processing, as the work gets there.

    • sensors
    • ISP
    • debayer
Rules

How we work

  1. We ship things

    Every project ends in something you can run.

  2. It's open

    Code is Apache-2.0, writing is CC BY 4.0, repos are public.

  3. We show our work

    If you can't rerun it, we haven't finished it.

  4. AI proposes, people approve

    A person makes the call on anything that changes a live system or an archive.

Name

Why “Mezzframe”

In a post house, the mezzanine file is the high quality master that sits between the camera original and every delivery copy. It is the version you go back to when you need to make something new.

That is the spot this lab wants to work in: the careful middle step where pictures are measured, cleaned up and handed on in better shape than they arrived.

  1. Camera original
  2. Mezzanine◉ We work here
  3. Delivery
Call in

Come say hi

Right now it's one broadcast engineer and a lot of side projects. We'd like it to grow into a group of video engineers, AI builders, archivists and curious people trading ideas, code and test results.

Start on GitHub