MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641105679 A) filed by Mr. Abhivanth R; Mr. Barath P; Mr. Loheth M; and Ms. Preetha P on September 02, 2026, for Self-Supervised Joint Embedding Predictive Architecture For Real-Time Anomaly And Defect Detection.
Inventors include Mr. Abhivanth R; Mr. Barath P; Mr. Loheth M; and Ms. Preetha P.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: Self-supervised joint embedding predictive architecture for real-time anomaly and defect detection Abstract: The present invention relates to the technical field of computer vision and machine learning, specifically self-supervised anomaly detection in video streams. Existing detection systems rely on extensively labeled datasets or struggle to adapt to dynamic environmental changes. The present invention advances the prior art by integrating a partially fine-tuned Vision Transformer with a Cross-Attention Spatial Joint Embedding Predictive Architecture (JEPA), multi-scale temporal transformers for short and long time-window evaluations, and a Deep Support Vector Data Description (SVDD) energy model. This self-supervised configuration eliminates the need for manual bounding boxes or labels by deriving learning signals directly from the data. Furthermore, an uncertainty-aware scoring module utilizing Monte Carlo dropout suppresses false positives. The principal use of this invention is the real-time detection of structural defects, anomalous objects, and operational drifts in industrial and exhibition environments by establishing robust normative representations from unlabeled video sequences.
Disclaimer: Curated by HT Syndication.
