MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641112814 A) filed by Srm Institute Of Science And Technology, Ramapuram Campus; and Easwari Engineering College on September 21, 2026, for Zero-Shot Semantic Segmentation Using Swin V2 Transformer With Masked Attention Guided Decoding.
Inventors include A Yassar Sharief; Niharika M; Subhiksha S K; V. Gowri; and Dr. D. Rajalakshmi.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: Abstract: The present invention discloses a computer-implemented language-free zero-shot semantic segmentation system and method for assigning pixel-level labels to previously unseen object categories in digital images. The system employs a Swin V2 Transformer backbone for hierarchical visual feature extraction, a Flash Masked Attention Decoder with multiple stacked layers for progressive foreground-focused attention refinement, and a Deformable Pixel Decoder for adaptive reconstruction of high-resolution segmentation outputs. The invention further utilizes cosine similarity-based matching between extracted pixel representations and visual prototypes of unseen classes for accurate semantic prediction without textual supervision. Experimental evaluation on the ADE20K dataset demonstrates improved segmentation accuracy, pixel-level classification performance, and boundary-aware segmentation capability, thereby enabling robust recognition and segmentation of unseen objects in complex visual environments.
Disclaimer: Curated by HT Syndication.
