V-JEPA
Meta's first Video Joint Embedding Predictive Architecture for self-supervised latent video learning.
DemonstratedCurrent
Approach
- Family
- Latent dynamics
- Weights
- Open weights
- API
- Unknown
- Product
- Unknown
Evidence-backed claim
Meta's V-JEPA learns predictive latent video features from unlabeled footage using a joint-embedding prediction objective.
Open primary source