Encoders¶
The model card fields and every encoder in v0.3.0. Encoders explains the encoder folder, images and the check.
Model card¶
Contract: model card, src/medomni/core/contracts/schemas/model-card.schema.json. One per encoder, at perception/encoders/<family>/<encoder>/card.yaml, and published to Hugging Face.
| Field | Required | Meaning |
|---|---|---|
id |
yes | <family>__<name>, matching the encoder's folder. |
family |
yes | The one family it reads. |
modality |
yes | The modality, in plain words. |
summary |
yes | One sentence on what it does. |
authors |
yes | Who built the encoder: name and GitHub handle for each. |
intended_use |
no | The population and acquisition it was built for. |
limits |
no | Known conditions where it is out of distribution. |
outputs |
yes | Each output block's kind (labels, events, text, embedding), with its scale, full labels vocabulary or dim where relevant. |
options |
no | Option name to description. |
reliability |
yes | status (unmeasured or measured) and optional evidence. |
non_training |
yes | Whether it runs zero-shot, with no training on MedOmni data. |
license |
yes | License of the code and weights. |
source |
yes | url and rev of the upstream code. |
weights |
no | uri and sha256 of the weights. |
runtime |
no | image and digest of the container. |
acceptance |
yes | pending or accepted, set by review. |
A model card has no extra keys, its family exists, and an encoder never returns a label outside its declared vocabulary.
Encoders in v0.3.0¶
44 encoders. Each is built as an image on your machine with medomni perception check; see Check and freeze.
Licenses differ. Some are non-commercial (CC BY-NC) or copyleft (GPL); read the model card before using an encoder outside research.
| Encoder | Family | License |
|---|---|---|
body-position-code-10hz__rule-occupancy |
body-position-code-10hz |
Apache-2.0 |
cgm-5min-mgdl__pyagata |
cgm-5min-mgdl |
GPL-3.0-only |
clinical-measurement-irregular__labs-summary |
clinical-measurement-irregular |
Apache-2.0 |
cxr-radiograph-2d__biomedclip |
cxr-radiograph-2d |
MIT |
cxr-radiograph-2d__chexzero |
cxr-radiograph-2d |
MIT |
cxr-radiograph-2d__torchxrayvision |
cxr-radiograph-2d |
Apache-2.0 |
ecg-12lead-500hz-mv__ecgfounder |
ecg-12lead-500hz-mv |
MIT |
ecg-1lead-200hz-uv__ecgfounder |
ecg-1lead-200hz-uv |
MIT |
ecg-1lead-200hz-uv__neurokit2 |
ecg-1lead-200hz-uv |
MIT |
ecg-monitor-3to5lead-250hz-continuous__ecgfounder |
ecg-monitor-3to5lead-250hz-continuous |
MIT |
eeg-sleep-6ch-200hz-uv__yasa |
eeg-sleep-6ch-200hz-uv |
BSD-3-Clause |
emg-chin-200hz-uv__rule-amplitude |
emg-chin-200hz-uv |
Apache-2.0 |
emg-limb-200hz-uv__rule-amplitude |
emg-limb-200hz-uv |
Apache-2.0 |
eog-2ch-200hz-uv__wav2sleep-eog |
eog-2ch-200hz-uv |
MIT |
flio-tcspc-256x256x1024bin-u16__rule-photons |
flio-tcspc-256x256x1024bin-u16 |
Apache-2.0 |
fundus-faf-mono8-2d__retinograd |
fundus-faf-mono8-2d |
MIT |
fundus-faf-mono8-2d__rule-intensity |
fundus-faf-mono8-2d |
Apache-2.0 |
fundus-photography-2d__dr-grade |
fundus-photography-2d |
MIT |
fundus-photography-2d__retfound |
fundus-photography-2d |
CC-BY-NC-4.0 |
fundus-photography-2d__retfound-dr |
fundus-photography-2d |
CC-BY-NC-4.0 |
fundus-photography-2d__retfound-jsiec |
fundus-photography-2d |
CC-BY-NC-4.0 |
oct-bscanset-radial24-circle3-mono8__mirage-onh |
oct-bscanset-radial24-circle3-mono8 |
CC-BY-NC-ND-4.0 |
oct-bscanset-radial24-circle3-mono8__rule-intensity |
oct-bscanset-radial24-circle3-mono8 |
Apache-2.0 |
oct-volume-raster-mono8-3d__kdoct |
oct-volume-raster-mono8-3d |
MIT |
oct-volume-raster-mono8-3d__mirage |
oct-volume-raster-mono8-3d |
CC-BY-NC-ND-4.0 |
oct-volume-raster-mono8-3d__rule-intensity |
oct-volume-raster-mono8-3d |
Apache-2.0 |
octa-enface-slab-mono8-2d__rule-intensity |
octa-enface-slab-mono8-2d |
Apache-2.0 |
octa-enface-slab-mono8-2d__vessel-density |
octa-enface-slab-mono8-2d |
Apache-2.0 |
octa-flowcube-mono8-3d__flow-metrics |
octa-flowcube-mono8-3d |
Apache-2.0 |
octa-flowcube-mono8-3d__rule-intensity |
octa-flowcube-mono8-3d |
Apache-2.0 |
ppg-waveform-uncal__bayesbeat |
ppg-waveform-uncal |
UNDECLARED |
ppg-waveform-uncal__neurokit2 |
ppg-waveform-uncal |
MIT |
pressure-nonarterial-125hz-mmhg__rule-summary |
pressure-nonarterial-125hz-mmhg |
Apache-2.0 |
psg-cardiorespiratory-4ch-200hz-mixed__wav2sleep |
psg-cardiorespiratory-4ch-200hz-mixed |
MIT |
pulse-rate-10hz-bpm__rule-summary |
pulse-rate-10hz-bpm |
Apache-2.0 |
resp-effort-rip-2ch-50hz__neurokit2 |
resp-effort-rip-2ch-50hz |
MIT |
resp-flow__rule-amplitude |
resp-flow |
Apache-2.0 |
resp-impedance-62hz-ohm__rule-amplitude |
resp-impedance-62hz-ohm |
Apache-2.0 |
retinal-heightmap-surfaces-f32__layer-thickness |
retinal-heightmap-surfaces-f32 |
Apache-2.0 |
retinal-heightmap-surfaces-f32__rule-surfaces |
retinal-heightmap-surfaces-f32 |
Apache-2.0 |
snore-acoustic__rule-amplitude |
snore-acoustic |
Apache-2.0 |
spo2-series-10hz-pct__pobm |
spo2-series-10hz-pct |
GPL-3.0 |
wearable-hr-1min-bpm__rule-summary |
wearable-hr-1min-bpm |
Apache-2.0 |
wearable-spo2-1min-pct__rule-summary |
wearable-spo2-1min-pct |
Apache-2.0 |
Where next¶
- Use the encoders: pull an image and call it.
- Gateway API: how cards and evidence are served.
- Families: the input each family expects.