Spotlight is a benchmark for evaluating whether VLMs can detect, localize, and explain fine-grained errors in high-fidelity text-to-video generations, with 600 videos and over 1,600 annotated error localizations.
ECCV 2026
SPIKE and SPIKE-RL use Bayesian Surprise to identify memorable moments in videos and guide surprise-weighted frame sampling, improving Video-LLM performance across downstream benchmarks.
ICLR 2026
We argue that world models need counterfactual simulation: the ability to reason across alternative realities and what-if scenarios, enabling safer, more capable, and more creative agents in novel environments.
NeurIPS 2025 LAW Workshop
The biases exhibited by text-to-image (TTI) models are often treated as independent, though in reality, they may be deeply interrelated. Fixing bias along one dimension can inadvertently affect another. We show measure and quantify such interactions, and propose InterMit, an intersectional bias mitigation algorithm that leverages these insights to achieves superior results with fewer steps.
EMNLP 2025
Black Swan is a benchmark for evaluating vision-language models' (VLMs) commonsense reasoning capabilities in videos, particularly in abductive and defeasible reasoning tasks. It focuses on atypical events, requiring models to reason about unexpected occurrences and adapt their hypotheses based on new information.
CVPR 2025
We propose a general approach to study and quantify a broad spectrum of biases, for any TTI model and for any prompt, using counterfactual reasoning. Unlike other works that evaluate generated images on a predefined set of bias axes, our approach automatically identifies potential biases that might be relevant to the given prompt, and measures those biases.
ECCV 2024
We introduce the GlobalRG benchmark, comprising two challenging tasks, retrieval across universals and cultural visual grounding, to test VL models' cultural inclusivity.
EMNLP 2024
We present a new Vision-Language-Commonsense transformer model, VLC-BERT, that incorporates contextualized knowledge using Commonsense Transformer (COMET) to solve Visual Question Answering (VQA) tasks that require commonsense reasoning.
WACV 2023
Developed DE-TensoRF, a model that can render 3D objects with as few as 3 images, and in under 15 min on a single GPU. We achieved the highest grade in our class, and led to collaboration efforts with Dr. Helge Rhodin’s research group.
Developed an extension to VisualCOMET to generate general-purpose commonsense knowledge from images. Showed improvements on coherence and diversity scores of a novel topic modelling algorithm using the generated knowledge
Incorporated Graph Neural Networks in a visual-linguistic Transformer
This project extends VLC-BERT with pointer generator networks and object detection models. Furthermore, we compare the performance of VLC-BERT with GPT-3 on the OK-VQA dataset.
This project introduces a modification to the GA algorithm to introduce dynamic mutation power, to solve the Lunar Lander evironment on OpenAI Gym in 30 generations.
In this project paper, we surveyed text summarization models by evaluating existing extractive and abstractive models. We studied the metrics and datasets used to evaluate the latest models and evaluated upcoming abstractive techniques. Finally, we highlighted future pathways for text summarization and suggested areas for improvement
A powerful Python API template, built on Flask, for plug-and-play use with machine learning models.