MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641058184 A) filed by Ms. P. Santhoshi; Mr. T. Krishnakumar; Mr. S. Prasanth; Mr. M. Mathubalan; and Mr. D. P. Ratheesb on May 07, 2026, for A System And Method For Image Recommendation And Semantic Segmentation Using A Clip Based Multimodal Learning Model With Adaptive Cross-Model Refinement.
Inventors include Mr. T. Krishnakumar; Mr. S. Prasanth; Mr. M. Mathubalan; and Mr. D. P. Ratheesb.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: ABSTRACT The present invention discloses a computer-implemented system and method for joint image recommendation and semantic segmentation using a multimodal CUP-based framework. The system. receives queries in image or textual format and maps them into a shared 512cdime.nsio.nal semantic feature space. A recommendation module ranks images using similat"ity scores, while ::1 segmentation· rl1odule generates pixel-level masks via embedding-guided attention. A core feature is the Adaptive Cross~Modal Refmement Meehanism, which iteratively improves seginentation precision through .similarity-weighted fusion of textual and visWtl embeddings: A feedback-driven learning module dynamically recalibrates QU!king pa!'affieters basecl on us.er interaction data. This bidiff!cJio~l· optimi~tion ensures high. context_ual relf!vanee and computational efficiency at;ross applications such a:s e-commerce; fuedieal imaging, and surveillance.
Disclaimer: Curated by HT Syndication.
