Publications

Agentic Very Long Video Understanding

Aniket Rege , Arka Sadhu , Yuliang Li , Kejie Li , Ramya Korlakai Vinayak , Yuning Chai , Yong Jae Lee , Hyo Jin Kim

Presents an agentic framework for very long video understanding that combines visual search, transcript search, and entity-graph reasoning.

Accepted to ACL 2026.

DiVE-k: Differential Visual Reasoning for Fine-Grained Image Recognition

Raja Kumar , Arka Sadhu , Ram Nevatia

Introduces a differential visual reasoning framework that uses preference-driven rollouts to improve fine-grained image recognition.

Published as a conference paper at ICLR 2026.

Vision-Language Pre-training Generalization: From Image-Text Pairs to Diverse Vision-Text Tasks

Arka Sadhu , Ram Nevatia

Studies how task-specific vision-language pre-training can transfer from image-text supervision into broader image and video reasoning tasks.

Presented at WACV 2024.

Unaligned Video-Text Pre-training using Iterative Alignment

Arka Sadhu , Licheng Yu , Animesh Sinha , Yu Chen , Ram Nevatia , Ning Zhang

Explores iterative alignment strategies for learning from video and text pairs that are only weakly or noisily aligned.

Gradient-based Memory Editing for Task-Free Continual Learning

Xisen Jin , Arka Sadhu , Junyi Du , Xiang Ren

Introduces memory editing mechanisms for task-free continual learning to better control forgetting without relying on explicit task boundaries.

Improving Object Detection and Attribute Recognition by Feature Entanglement Reduction

Zhaoheng Zheng , Arka Sadhu , Ram Nevatia

Investigates how disentangling feature representations can jointly improve object detection and attribute recognition.

Utilizing Every Image Object for Semi-supervised Phrase Grounding

Haidong Zhu , Arka Sadhu , Zhaoheng Zheng , Ram Nevatia

Shows how object-level structure can improve semi-supervised phrase grounding by making fuller use of image content.

Video Question Answering with Phrases via Semantic Roles

Arka Sadhu , Kan Chen , Ram Nevatia

Connects phrase-level supervision and semantic-role structure to improve video question answering beyond surface text matching.

Visual Semantic Role Labeling for Video Understanding

Arka Sadhu , Tanmay Gupta , Mark Yatskar , Ram Nevatia , Aniruddha Kembhavi

Presents a semantic-role-centered formulation for video understanding, pairing structured roles with action-centric video reasoning.

Built into the VidSitu benchmark and project site.

Video Object Grounding using Semantic Roles in Language Description

Arka Sadhu , Kan Chen , Ram Nevatia

Grounds video objects through the semantic structure of language, tying textual roles to video evidence over time.

Visually Grounded Continual Learning of Compositional Phrases

Xisen Jin , Junyi Du , Arka Sadhu , Ram Nevatia , Xiang Ren

Combines visually grounded language learning with continual learning, focusing on compositional phrase understanding over time.

Zero-Shot Grounding of Objects from Natural Language Queries

Arka Sadhu , Kan Chen , Ram Nevatia

Addresses phrase grounding in the zero-shot setting, focusing on transferring to unseen queries and concepts.

Accepted as an oral presentation at ICCV 2019.