IMAGEGAFTER: AI-DRIVEN DATA AUGMENTATION FOR COMPUTER VISION
Madhuri Barhate, Rutuja Diwate, Hitesh Tolani, Diksha Teware, Ashwanti Gaikwad, Shravan Kalzunkar
Vishwakarma Institute of Technology, Pune, India
ABSTRACT: ImageGafter introduces a Python-based software that leverages large language models (LLMs) to address the persistent issue of image scarcity in large-scale deep learning models. The software takes an input image and uses advanced algorithms to generate coherent and descriptive text representations of the visual content. By seamlessly integrating cutting-edge AI technologies, it bridges the gap between visual and textual data, tackling the challenge of limited annotated images in deep learning. The project goes beyond text generation by creating a diverse set of prompts using Generative AI. These prompts are designed to capture various features of the input image, and are then used by Generative AI to create new images, effectively expanding the dataset. This artificial dataset growth provides a richer source of information for training deep learning models. By combining LLM-driven image-to-text generation with prompt-based image synthesis, the project offers an innovative solution to image scarcity, paving the way for more robust and accurate deep learning models.
KEYWORDS: Generative AI, Graphical user interface, LLM, Prompt generation, Image generation.