metadata
pipeline_tag: image-feature-extraction
MeFEm: Medical Face Embedding model
MeFEm is a vision model based on a modified Joint Embedding Predictive Architecture (JEPA) for biometric and medical analysis from facial images.
Model Description
MeFEm introduces several modifications to the JEPA framework to optimize for medical and biometric facial analysis:
- Axial stripe masking strategy: Focuses learning on semantically relevant regions of the face.
- Circular loss weighting scheme: A novel weighting approach for the training objective.
- Probabilistic reassignment of the CLS token: Designed to improve the quality of linear probing for downstream tasks.
The model was trained on a consolidated dataset of curated images and outperforms strong baselines like FaRL and Franca on core anthropometric tasks and Body Mass Index (BMI) estimation, despite using significantly less data.
Resources
Citation
@article{mefem2026,
title={MeFEm: Medical Face Embedding model},
author={},
journal={arXiv preprint arXiv:2602.14672},
year={2026}
}