Intellectual Property India Publishes Patent Application for 'Household Object Visual Question Answering: Evaluation of Zero-Shot and Fine-Tuned Vision-Language Models' Filed by Tatikonda Krishna Chaitnya; K. V. S. Sri Harsha; Bhuma Chandra Mohan; Naga Raju Challa; Naralasetty Mounika; Koneti Surya Mounika; Medisetti Harsha Vardhan; Kurmala Siva Kumar; and Bapatla Engineering College

MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641048408 A) filed by Tatikonda Krishna Chaitnya; K. V. S. Sri Harsha; Bhuma Chandra Mohan; Naga Raju Challa; Naralasetty Mounika; Koneti Surya Mounika; Medisetti Harsha Vardhan; Kurmala Siva Kumar; and Bapatla Engineering College, Bapatla, Andhra Pradesh, on April 16, for 'household object visual question answering: evaluation of zero-shot and fine-tuned vision-language models.'

Inventor(s) include K. V. S. Sri Harsha; Bhuma Chandra Mohan; Naga Raju Challa; Tatikonda Krishna Chaitanya; Naralasetty Mounika; Koneti Surya Mounika; Medisetti Harsha Vardhan; and Kurmala Siva Kumar.

The application for the patent was published on May 1, under issue no. 18/2026.

According to the abstract released by the Intellectual Property India: "Recognizing household objects by using Vision-Language Models (VLMs) is an important step towards building intelligent systems for facilitating communication and interaction. Household objects are of special interest because of their varied nature of shape, color, texture, and use. In this regard, we propose a novel Visual Question Answering (VQA) dataset of 1,000 image-question pairs of household objects. In our proposed dataset, various attributes of object recognition and features are included. The proposed method has the potential for building VLMs. To assess the performance of various state-of-the-art VLMs, we propose zero-shot and fine-tuning evaluation approaches. In our zero-shot evaluation method, we use Ristretto-3B models. We achieve a mean cosine similarity of 76%. This shows the potential of various VLMs for object recognition and understanding of objects at home. In our proposed fine-tuning evaluation method, we use Qwen2-VL-2B models. We achieve a mean cosine similarity of 84%. We use various platforms such as Google Colab and Kaggle for training our models. This shows the potential of our proposed fine-tuning method for object recognition and understanding of objects at home. The potential of VLMs for object recognition and understanding of objects at home has various applications in assistive technology for the visually impaired, robotics, smart home technology, augmented reality, etc."

Disclaimer: Curated by HT Syndication.