Face expression matcher (linear-rank-v1)

Matches ten expressions to the ten photos of one person from 78 Py-Feat features plus their within-person ranks.

  • trained on 66 unique people (660 photos), duplicates across uploads removed
  • model: ens_ridge1000_lr0.01
  • nested CV matching accuracy on unseen people: 0.699 (chance 0.10)
  • weakest pairs: sad vs hopelessness, scepticism vs uncertainty

Files: model.json (grader format), model.safetensors + config.json (used by the Streamlit app), metrics.json, reports/. Built by Gaurav Tomar.

Downloads last month
39
Safetensors
Model size
1.56k params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Space using 23f1000805/face-expression-v3 1