I'm deeply fascinated by the synergy between different types of data including text, images, videos, audio, and more, mirroring a subset of human senses. Eventually, I want to work at this big intersection. As a stepping stone towards this goal, I am currently studying the interaction between vision and language and robot actions.
News & Updates
Our paper on dual-stage efficient token reduction for VLMs has been accepted at IEEE/CVF CVPR 2026.
PaperConducted a hands-on workshop on fine-tuning AI models using Unsloth at IIT Delhi.
WorkshopConducted a workshop on fine-tuning AI models with Meta at IIT Bombay.
WorkshopConducted a workshop on fine-tuning AI models with Meta at IISc Bangalore.
WorkshopJoined AMD to work on efficiency and performance of diffusion models, LLMs, VLMs and VLAs.
MilestoneOur paper on motion-guided diffusion for GIF generation has been accepted at the European Conference on Computer Vision (ECCV) 2024.
PaperPix2Gif was accepted at the AI for Content Creation (AI4CC) Workshop at CVPR 2024, Seattle.
PaperPublications
DUET-VLM: Dual-Stage Unified Efficient Token Reduction for VLM Training and Inference
Aditya Kumar Singh*, , Pratik Prabhanjan Brahma, Zicheng Liu, Emad Barsoum
(* Equal Contribution)
CVPR 2026 | IEEE/CVF Conference on Computer Vision and Pattern Recognition
pdf
abstract
code
cite
Beyond Boundaries: A Novel Data-Augmentation Discourse for Open Domain Generalization
Shirsha Bose, Ankit Jha, , Biplab Banerjee
TMLR | Transactions on Machine Learning Research
paper
cite
Multi-Stage Semantic Graph Embeddings for Compositional Zero-Shot Learning
, Ruchika Chavhan, Ushasi Chaudhuri, Biplab Banerjee
paper
cite
