V-JEPA
PulseAugur coverage of V-JEPA — every cluster mentioning V-JEPA across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
STATERA model estimates hidden mass from video using frozen temporal representations
Researchers have developed STATERA, a novel method for estimating the center-of-mass (CoM) of opaque, asymmetric rigid bodies from short monocular videos. This approach adapts a pretrained video backbone, V-JEPA, using …
-
AI City Challenge 2026: New framework wins with decoupled semantic understanding
Researchers have developed a novel framework for traffic scene understanding that decouples semantic fact extraction from natural language generation, addressing issues of hallucination and inconsistent reasoning in exi…
-
Lung ultrasound AI research compares self-supervised learning methods
A new research paper explores the effectiveness of different self-supervised learning (SSL) pretext tasks for lung ultrasound (LUS) image analysis. The study compared contrastive learning (MoCo), masked reconstruction (…
-
New Arti-JEPA model adapts video models for vocal tract MRI analysis
Researchers have developed Arti-JEPA, a new joint embedding predictive architecture designed to model real-time MRI data of the vocal tract for speech analysis. This model was trained on approximately 62 hours of unlabe…
-
New ViTAMINS method enhances vision transformer training with synthetic negatives
Researchers have developed ViTAMINS, a novel method for training self-supervised vision transformers by incorporating synthetic hard negatives. This approach enhances representation quality, leading to significant impro…
-
New video pretraining method and 10M-hour dataset released
Researchers have introduced LeVJEPA, a novel video pretraining method that significantly reduces computational costs while maintaining or improving downstream accuracy. This approach bypasses common heuristics like arch…
-
New models for autonomous driving predict future world states and actions · 2 sources tracked
Researchers have developed new models for autonomous driving that focus on predicting future world states and actions. WA-JEPA, presented in one paper, adapts the Video Joint Embedding Predictive Architecture (V-JEPA) b…
-
Gemini 2.5-Flash leads multimodal rapport estimation in real-world HRI study
A new arXiv paper explores multimodal rapport estimation in real-world Human-Robot Interaction (HRI) settings, moving beyond controlled lab environments. Researchers found that zero-shot Large Language Models (LLMs) per…
-
New V-JEPA4A model enhances autonomous driving video analysis
Researchers have developed V-JEPA4A, a new self-supervised learning model specifically designed for autonomous driving applications. This model utilizes a novel saliency-driven masking policy, which prioritizes semantic…
-
New benchmark H2R-Bench reveals limitations in human-to-robot video generation
Researchers have introduced H2R-Bench, a new benchmark designed to evaluate video generation models' ability to translate human manipulation videos into robot-centric demonstrations. The benchmark addresses the challeng…
-
JEPA models fail at novel driving data detection due to domain shift
A new research paper titled "Asleep at the Wheel: JEPA's Limitations in Evaluating Novel Driving Data" published on arXiv highlights a critical flaw in using self-supervised learning models, specifically Joint-Embedding…
-
Deep learning framework estimates coastal wave parameters from video
Researchers have developed a novel deep learning framework for estimating five key coastal wave parameters from monocular video. This system utilizes a V-JEPA backbone for feature extraction in challenging visual condit…
-
University of Michigan unveils NeuroVFM for neuroimaging analysis
Researchers at the University of Michigan have developed NeuroVFM, a novel foundation model for neuroimaging. Trained using the Vol-JEPA approach on over 5.24 million clinical MRI and CT scans, NeuroVFM learns from uncu…
-
New JEPA-style model learns useful network fingerprint embeddings
Researchers have developed JA4-JEPA, a Transformer-based model that applies JEPA-style predictive learning to network fingerprints. This approach, which learns by matching latent predictions rather than regenerating inp…
-
AI framework enhances collision prediction in transport systems
Researchers have developed a novel spatiotemporal semantic V2X framework designed to improve collision prediction in intelligent transportation systems. This framework utilizes the Video Joint Embedding Predictive Archi…
-
AI needs world models for real-world tasks, JEPA shows promise over LLMs
Pascale Fung, a leading researcher in world models, presented at ICML 2026 on the necessity of world models for AI agents operating in the real world. She argued that while Large Language Models (LLMs) and Vision-Langua…
-
SiamJEPA uses Siamese encoders for improved self-supervised learning
Researchers have introduced SiamJEPA, a novel approach to self-supervised representation learning that utilizes Siamese student encoders within Joint Embedding Predictive Architectures (JEPAs). Unlike previous JEPA mode…
-
Drive-JEPA framework advances end-to-end autonomous driving with novel video pretraining
Researchers have introduced Drive-JEPA, a novel framework that combines Video Joint-Embedding Predictive Architecture (V-JEPA) with multimodal trajectory distillation for end-to-end autonomous driving. This approach ada…
-
New framework learns implicit 3D physics from video
Researchers have developed a self-supervised framework called Neural Voxel Dynamics that learns implicit 3D physics directly from video. This method addresses limitations in current generative video models by predicting…
-
Chinese AI startup Fysics AI launches physics-based world model
Shanghai-based Fysics AI has launched Fysiverse, a new AI world model that incorporates real-world physical laws directly into its code. This approach differs from the data-driven methods used by companies like OpenAI a…