MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641125342 A) filed by Vellore Institute Of Technology on October 21, 2026, for Segment–inpaint–cross-Class Recompose Pipeline For Augmentation Of Classroom-Based Waste Image Generation For Classification.
Inventors include Arpan Garai; and Diya.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: ABSTRACT The present invention relates to a computer-implemented system and method for generating a synthetic training dataset for automated classroom waste classification. The invention provides a Segment–Inpaint–Cross-Class Recompose (SICR) pipeline comprising segmentation (3) of a waste object from a source frame to generate a foreground cut-out, inpainting (4) of an object region in a different-class frame to reconstruct an object-free background, and cross-class recomposition (5) of the foreground object onto the reconstructed background. The resulting composite image is assigned a class label corresponding to the pasted waste object rather than the background class, thereby reducing contextual bias and encouraging object-specific feature learning. Multiple cross-class backgrounds are paired with a limited set of curated foregrounds to multiplicatively generate synthetic samples for an augmented training pool (6). A video-level split protocol prevents frames from a common recording from being distributed across training, validation, and testing sets. The resulting dataset is used for training a lightweight classifier (7), including MobileNetV2, to generate classification output (8) for paper, plastic, rubber, and biodegradable waste.
Disclaimer: Curated by HT Syndication.
