What is CMF?

The Common Model Format (CMF) is a packaging standard that lets any detection model be loaded and run inside RIME — without custom integration code.

The problem CMF solves

Today, each lab's FOG detection model is a standalone script. It reads data in one format, produces output in another, and cannot be connected to an annotation tool without significant engineering work.

flowchart TD
    A[Lab A annotates video] -->|manual scripts| B[Lab A trains FOG detector]
    B -->|standalone model| C[Cannot load into annotation tool]
    C -->|Lab B wants to test it| D[Lab B writes custom integration]
    D --> E[Lab B reformats data]
    E --> F[Lab B interprets output manually]

The result: models accumulate in papers but cannot be compared, reused, or evaluated against gold-standard annotations without significant effort.

The CMF contract

A CMF package declares:

1. What it needs — which signal channels, at what sampling rate; or video
2. What it produces — probabilities, event intervals, or point events
3. Where to display output — which annotation lane and label
4. How to run inference — sliding window or whole-signal
5. What parameters the user can adjust — thresholds, window sizes

flowchart LR
    subgraph pkg [".rime package"]
        cfg[config.json]
        wrap[wrapper.py]
    end
    RIME -->|reads| cfg
    RIME -->|runs| wrap
    cfg -->|tells RIME| i["Inputs: channels, Hz, shape"]
    cfg -->|tells RIME| o["Outputs: probability / intervals / points"]
    cfg -->|tells RIME| d["Display: lane + label"]

What a .rime package looks like

freeze-index.rime/
├── config.json      # the contract
└── wrapper.py       # the model

Any model that implements this contract can be loaded by RIME and run on any compatible session — no integration work required.

See Loading a Model to get started.