Projects
Research & engineering across computer vision, generative models, and robotics.
3D Reconstruction from SPAD Camera Data
Leveraging radiance fields with SPAD photon-counting datasets as priors for high-quality 3D reconstructions in challenging low-light conditions.
Night Time Video Flare Removal
Deep learning pipeline for removing nighttime lens flare from videos using a large-scale synthetically generated dataset and diffusion-based architectures.
Virtual Try-Ons for Indic Clothing
GAN and diffusion-based virtual try-on system specialised for Indian traditional clothing, built on a custom scraped and annotated dataset.
Speech Disfluency Detection
Custom ResNet classifier trained on mel-spectrograms that detects filler words and repetitions in speech with 75 % accuracy, deployed on Heroku.
Sign Language Video Generator using GAN
Government-funded (₹4 M) conditional GAN that synthesises Indian Sign Language pose-sequence videos from text glosses for accessibility applications.