IIT Roorkee sets milestone in AI & creative design with FLORA & NeRA adapter
PIONEER EDGE NEWS SERVICE/Dehradun
The Indian Institute of Technology (IIT) Roorkee claims to have achieved a milestone in artificial intelligence and creative design with the development of the FLORA dataset and NeRA adapter, two innovations that redefine the way AI understands, interprets and generates fashion. The study, titled “Dressing the Imagination: A Dataset for AI-Powered Translation of Text into Fashion Outfits and a Novel NeRA Adapter for Enhanced Feature Adaptation,” showcases India’s-first framework that enables AI models to generate detailed, high-fidelity outfit sketches from their textual descriptions.
Fashion thrives on imagination but designers have to spend a lot of time in translating creative ideas or concepts into sketches. Fashion designers also need to adapt to the latest fashion trends and propose personalised apparel designs. In the past, customers would search for a dress using simple keywords such as a “blue shirt”. However, modern customers and designers work with complex vocabulary such as “a minimalist, cropped rayon blouse with a Mandarin collar and raglan sleeves”. Many current AI tools struggle to understand this kind of professional fashion language. This research helps solve that problem by teaching AI to understand detailed clothing terms and translate them into fashion sketches.
The research, led by Sparsh Mittal of the department of Electronics and Communication Engineering, joint faculty in the institute’s Mehta Family School of Data Science and Artificial Intelligence, represents a major leap in India’s pursuit of AI-driven, precise and expressive design in the fashion industry.
Most existing fashion datasets rely on photographs or catalog images. These are helpful for shopping websites, but they don’t capture what designers usually draw in their sketchbooks. FLORA is different as it includes pairs of professional fashion sketches and their textual descriptions, using the kind of terms designers use, such as silhouettes, materials, textures, and overall style. When an AI model is trained on the FLORA dataset, it can learn to connect detailed words with the right visual features.
Along with FLORA, the NeRA (Nonlinear Expressive Representation Adapter) adapter is a new method that helps AI models learn these design connections faster and more accurately. The open-source release of FLORA and NeRA will empower researchers and designers globally to collaborate and innovate at the intersection of art, technology and sustainability.
Mittal said, “Our work brings AI and design together to make creativity easier, faster, and more sustainable. Modern fashion includes small details that many AI models could not understand well. With FLORA and NeRA, we help AI understand a designer’s ideas through detailed sketches and precise design terms. This can speed up new developments in fashion, retail and other areas.”
IIT Roorkee director Kamal Kishore Pant said, “These innovations like FLORA and NeRA demonstrate how AI can be harnessed to enhance India’s creative economy and bolster its global leadership in design and technology.”
