Sanjoy Chowdhury

I am a third year CS PhD student at University of Maryland, College Park advised by Prof. Dinesh Manocha. I am broadly interested in multi-modal learning and its different applications. My research primarily involves studying the interplay between the vision and audio modalities and developing systems equipped with their comprehensive understanding.

I am currently working as a ML Research intern at Apple MLR hosted by Chun-Liang Li and Karren Yang . I spent the summer of '24 at Meta Reality Labs working as a research scientist intern hosted by Ruohan Gao . Before this, I was a student researcher at Google Research with Avisek Lahiri and Vivek Kwatra in the Talking heads team on speech driven facial synthesis. Previously, I spent a wonderful summer with Adobe Research working with Joseph K J in the Multi-modal AI team as a research PhD intern on multi-modal audio generation. I am also fortunate to have had the chance to work with Prof. Kristen Grauman , Prof. Salman Khan , Prof. Mohamed Elhoseiny among other wonderful mentors and collaborators.

Before joining for PhD, I was working as a Machine Learning Scientist with the Camera and Video AI team at ShareChat, India. I was also a visiting researcher at the Computer Vision and Pattern Recognition Unit at Indian Statistical Institute Kolkata under Prof. Ujjwal Bhattacharya. Even before, I was a Senior Research Engineer with the Vision Intelligence Group at Samsung R&D Institute Bangalore. I primarily worked on developing novel AI-powered solutions for different smart devices of Samsung.

I received my MTech in Computer Science & Engineering from IIIT Hyderabad where I was fortunate to be advised by Prof. C V Jawahar. During my undergrad, I worked as a research intern under Prof. Pabitra Mitra at IIT Kharagpur and the CVPR Unit at ISI Kolkata.

Email  /  GitHub  /  Google Scholar  /  LinkedIn  /  Twitter

Updates

Selected publications

I am interested in solving computer vision, computer audition, and machine learning problems and applying them to broad AI applications. My research focuses on applying multi-modal learning (Vision + X) for generative modeling and holistic cross-modal understanding with minimal supervision. Representative papers are highlighted.

AVTrustBench: Assessing and Enhancing Reliability and Robustness in Audio-Visual LLMs

We introduce Audio-Visual Trustworthiness assessment Benchmark (AVTrustBench), comprising 600K samples spanning over 9 meticulously crafted tasks, evaluating the capabilities of AVLLMs across three distinct dimensions: Adversarial attack, Compositional reasoning, and Modality-specific dependency. Using our benchmark we extensively evaluate 16 state-of-the-art AVLLMs. The findings reveal that the majority of existing models fall significantly short of achieving human-like comprehension, offering valuable insights for future research directions. To alleviate the limitations in the existing approaches, we further propose a robust, model-agnostic calibrated audio-visual preference optimization based training strategy CAVPref, obtaining a gain up to 30.19% across all 9 tasks.

Meerkat: Audio-Visual Large Language Model for Grounding in Space and Time

We present Meerkat, an audio-visual LLM equipped with a fine-grained understanding of image and audio both spatially and temporally. With a new modality alignment module based on optimal transport and a cross-attention module that enforces audio-visual consistency, Meerkat can tackle challenging tasks such as audio referred image grounding, image guided audio temporal localization, and audio-visual fact-checking. Moreover, we carefully curate a large dataset AVFIT that comprises 3M instruction tuning samples collected from open-source datasets, and introduce MeerkatBench that unifies five challenging audio-visual tasks.

Towards Determining Perceived Human Intent for Multimodal Social Media Posts using The Theory of Reasoned Action

We propose Intent-o-meter, a perceived human intent prediction model for multimodal (image and text) social media posts. Intent-o-meter models ideas from psychology and cognitive modeling literature, in addition to using the visual and textual features for an improved perceived intent prediction.

Can LLM’s Generate Human-Like Wayfinding Instructions? Towards Platform-Agnostic Embodied Instruction Synthesis

We present a novel approach to automatically synthesize “wayfinding instructions" for an embodied robot agent. In contrast to prior approaches that are heavily reliant on human-annotated datasets designed exclusively for specific simulation platforms, our algorithm uses in-context learning to condition an LLM to generate instructions using just a few references.

MeLFusion: Synthesizing Music from Image and Language Cues using Diffusion Models ( Highlight, Top 2.8% )

We propose MeLFusion, a model that can effectively use cues from a textual description and the corresponding image to synthesize music. MeLFusion is a text-to-music diffusion model with a novel "visual synapse", which effectively infuses the semantics from the visual modality into the generated music. To facilitate research in this area, we introduce a new dataset MeLBench, and propose a new evaluation metric IMSM.

APoLLo : Unified Adapter and Prompt Learning for Vision Language Models

Our method is designed to substantially improve the generalization capabilities of VLP models when they are fine-tuned in a few-shot setting. We introduce trainable cross-attention-based adapter layers in conjunction with vision and language encoders to strengthen the alignment between the two modalities.

AdVerb: Visually Guided Audio Dereverberation

We present a novel audio-visual dereverberation framework that uses visual cues in addition to the reverberant sound to estimate clean audio.

Measured Albedo in the Wild: Filling the Gap in Intrinsics Evaluation

In order to comprehensively evaluate albedo, we collect a new dataset, Measured Albedo in the Wild (MAW), and propose three new metrics that complement WHDR

AudViSum: Self-Supervised Deep Reinforcement Learning for Diverse Audio-Visual Summary Generation

Introduced a novel deep reinforcement learning-based self-supervised audio-visual summarization model that leverages both audio and visual information to generate diverse yet semantically meaningful summaries.

V-DESIRR: Very Fast Deep Embedded Single Image Reflection Removal

We have proposed a multi-scale end-to-end architecture for detecting and removing weak, medium, and strong reflections from naturally occurring images.

Listen to the Pixels

In this study, we exploited the concurrency between audio and visual modalities in an attempt to solve the joint audio-visual segmentation problem in a self-supervised manner.

Blog(s)

Have tried my hand at writing technical blogs.

The devil is in the details: Video Quality Enhancement Approaches

The blog contextualizes the problem of video enhancement in present-day scenarios and talks about a couple of interesting approaches to handle this challenging task.

Academic services

I have served as a reviewer for the following conferences:

CVPR: 2023, '24, '25

ICCV: 2023, '25

ECCV: 2024

NeurIPS: 2024

AAAI: 2025

WACV: 2022, '23, '24

ACMMM: 2023, '24

ACL: 2024

Template credits: Jon Barron and thanks to Richa for making this.