Shypd.aiThe largest open index of AI toolsjepa is an Open Source AI tool that provides PyTorch code and models for V-JEPA self-supervised learning from video. It enables learning visual representations from video without manual labels.
Trending
About
jepa is the official PyTorch codebase for V-JEPA (Video Joint Embedding Predictive Architecture), a method developed by Meta AI Research, FAIR, for self-supervised learning of visual representations from video. This tool allows users to train models by passively watching video pixels, producing versatile visual representations that perform well on downstream video and image tasks without model parameter adaptation. V-JEPA pretraining relies solely on an unsupervised feature prediction objective, avoiding the need for pretrained image encoders, text, negative examples, human annotations, or pixel-level reconstruction. It includes a model zoo with pretrained models and attentive probes for various tasks like K400, SSv2, ImageNet1K, Places205, and iNat21.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
