Ayush Gupta

I am a Ph.D. student at Johns Hopkins University in the department of Computer Science. I am advised by Prof. Rama Chellappa working on problems in Computer Vision and Machine Learning. I am supported by an IARPA grant, BRIAR. I am currently working on problems related to fine-grained computer vision and metric learning.

Previously I obtained a B.E. in Computer Science from Birla Institute of Technology and Science (BITS), Pilani. At BITS Pilani I was working under the guidance of Prof. Poonam Goyal on generating natural language descriptions of videos, and Dr. Yogesh S Rawat on Gait Recognition.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

• 10/2024: Our paper MimicGait: A Model-Agnostic Approach for Occluded Gait Recognition using Correlational Knowledge Distillation has been accepted to WACV 2025!
• 09/2024: Our paper GaitContour: Efficient Gait Recognition based on a Contour-Pose Representation has been accepted to WACV 2025!
• 05/2024: Starting a summer internship at SRI! I will be working on vision-language models.
• 04/2024: Our paper "Tackling Domain Shifts in Person Re-Identification: A Survey and Analysis" has been accepted at the CVPR 2024 Continual Learning Workshop!
• 10/2023: Our paper You Can Run but not Hide: Improving Gait Recognition with Intrinsic Occlusion Type Awareness is accepted as an oral paper at WACV 2024! See you in Hawaii!
• 08/2023: Starting a Teaching Assistantship in the Machine Perception course for Fall 2023
• 08/2022: Joined the AIEM Lab as a PhD student at Johns Hopkins University! I will be working on BRIAR, an IARPA grant.
• 05/2021: Started collaborating with CRCV Lab at UCF for undergraduate thesis
• 09/2020: Started working with the Language Technology Group at University of Hamburg on the CLARIN COVID Disinformation Hackathon
• 05/2020: Started internship at Indian Institute of Remote Sensing, ISRO

I am interested in all things computer vision and machine learning. Specifically, my current research interests are in Multimodal LLMs, where I am working on making LLMs see the world like humans do. I also work on Gait Recognition and Person Re-ID, on vision models which can identify people from long-range, turbulent, in-the-wild data.

We use multimodal LLMs for temporal grounding of question-answer pairs in unconstrained videos.

We use an auxiliary occlusion detector to solve the occlusion problem in long range gait recognition.

We tackle the occlusion problem in gait recognition using correlational knowledge distillation.

We develop a novel, efficient contour-based representation for gait recognition.

Predicting the Ejection Fraction from ultrasound images of the heart, utilizing the Segment Anything Model.

A survey on domain shift in Person Re-ID.

Developing a novel technique for unsupervised gait recognition using temporal self similarity.

An approach for generating natural language captions of videos using external knowledge bases.

Template Credits : Jon Barron