Facial Expressions Corpus

1,200 consent-cleared facial-expression portraits, FACS-annotated across seven emotion categories.

A balanced corpus of posed and spontaneous facial expressions for affect and expression research. Every subject signed a research-and-commercial release; each image carries emotion labels, FACS action-unit codes, and demographic metadata to support fairness auditing.

Dataset details

Supported tasks

• Facial-expression and emotion-recognition training
• Expression-transfer and avatar-animation models
• Fairness / bias auditing across demographic slices

Dataset structure

Each item ships as one aligned portrait plus a JSON sidecar with labels and landmarks:

{
  "id": "fe_00912",
  "image": "fe_00912.jpg",
  "emotion": "surprise",
  "facs_action_units": [1, 2, 5, 26],
  "intensity": 0.72,
  "landmarks_68": "fe_00912_landmarks.json",
  "demographics": { "age_band": "25-34", "self_reported_gender": "female" },
  "consent": "research+commercial"
}

Samples

Curation

Recruited participants performed prompted expressions and reacted to neutral stimuli in a controlled studio. Two annotators independently coded FACS action units; disagreements were adjudicated by a third. Demographic balance was targeted at recruitment, not corrected after the fact.

Personal and sensitive information

Contains identifiable human faces. Every subject provided informed written consent for research and commercial ML use, with the right to withdraw; withdrawn subjects are purged on the next release. Do not use to identify or surveil individuals.

Considerations & limitations

• Posed expressions over-represent extremes; subtle micro-expressions are under-sampled.
• Demographic coverage is broad but not census-representative — audit before deploying.
• Studio lighting is uniform; in-the-wild robustness is not guaranteed.

License & provenance

Collected under informed consent for research and commercial ML use. The sample archive (full_dataset.zip) is a free 50-item preview; the full set of 1,200 items unlocks with a purchase. Redistribution of the raw dataset is not permitted under either license.

Citation

@misc{aria-forge_0000000f,
  title  = {Facial Expressions Corpus},
  author = {Aria Forge Studio (@aria-forge)},
  year   = {2026},
  note   = {TensorForge marketplace dataset, LST-0000000F-0000-4000-8000-00000000000F}
}