Apoorva Beedu

I am a PhD candidate at Georgia Institute of Technology advised by Dr. Irfan Essa and Dr. Justin Romberg.

My research interests are primarily focused on understanding actions and object's states in videos, and have previously worked on 6D object pose estimation and human activity recognition. Broadly, my research is in the union of

Video Understanding + Multi-Modal Learning + Foundation Models

During the summer of 2021, I was an intern at Facebook Reality Labs, collaborating with Chengde Wan and Robert Wang. In 2020, I undertook an internship at Microsoft Research with Dr. Amol Ambadekar. In the summers of '18 and '17, I interned at a startup called NodeIn, working with Dr. Suresh Kannan. Prior to joining Georgia Institute of Technology, I spent a year at Asteria Aerospace Limited as a software Engineer.

Currently looking for full-time positions starting Spring 2025.

Email: abeedu3 [at] gatech [dot] edu

GitHub  /  Google Scholar  /  LinkedIn  /  CV

profile photo

News

Research

I'm interested in computer vision, machine learning, video analysis and multi-modal representations.

project image

Mamba Fusion: Learning Actions Through Questioning


Apoorva Beedu*, Zhikang Dong*, Jason Sheinkopf, Irfan Essa
ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
pdf / code / bibtex / project page /

We introduce MambaVL, an efficient mamba based fusion method for vision language models.

project image

Text Descriptions of Actions and Objects Improve Action Anticipation


Apoorva Beedu, Harish Haresamudram, Irfan Essa
ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
pdf / bibtex / project page / supp /

Introduces contrastive learning for Action Anticipation and evaluates using text data as additional modalities. Detailed technical report can be found here

project image

Limitations in Employing Natural Language Supervision for Sensor-Based Human Activity Recognition--And Ways to Overcome Them


Harish Haresamudram, Apoorva Beedu, Mashfiqui Rabbi, Sankalita Saha, Irfan Essa, Thomas Ploetz
Proceedings of the AAAI Conference on Artificial Intelligence 2025
pdf / bibtex / project page /

We investigate whether natural language supervision can be used for wearable sensor based Human Activity Recognition (HAR), and discover that surprisingly it performs substantially worse than standard end-to-end training and selfsupervision.

project image

Multimodal Contrastive Learning with Hard Negative Sampling for Human Activity Recognition


Hyeongju Choi, Apoorva Beedu, Irfan Essa
ICCV 2023 workshop on PerDream: PERception, Decision making and REAsoning through Multimodal foundational modeling
pdf / bibtex /

Introduces contrastive learning with hard negative sampling for Human Activity Recognition.

project image

Multi-Stage Based Feature Fusion of Multi-Modal Data for Human Activity Recognition


Hyeongju Choi, Apoorva Beedu, Harish Haresamudram, Irfan Essa
ArXiv preprint, 2022
pdf / bibtex /

We present a multi-modal framework that learns to effectively combine features from RGB Video and IMU sensors for Human Activity Recognition.

project image

End-to-End Multimodal Representation Learning for Video Dialog


Huda Alamri, Anthony Bilic, Michael Hu, Apoorva Beedu, Irfan Essa
NeurIPS 2022 workshop on Vision Transformers: Theory and Applications
pdf / bibtex /

We present a framework for video based dialog task.

project image

Video based Object 6D Pose Estimation using Transformers


Apoorva Beedu, Huda Alamri, Irfan Essa
NeurIPS 2022 workshop on Vision Transformers: Theory and Applications
pdf / code / bibtex /

We introduce a Transformer based 6D Object Pose Estimation framework VideoPose, comprising an end-to-end attention based modelling architecture, that attends to previous frames in order to estimate accurate 6D Object Poses in videos.

project image

Masked reconstruction based self-supervision for human activity recognition


Harish Haresamudram,Apoorva Beedu, Varun Agrawal, Patrick L Grady, Irfan Essa, Judy Hoffman, Thomas Plötz
Proceedings of the 2020 ACM International Symposium on Wearable Computers
pdf / bibtex /

We introduce masked reconstruction as a viable self-supervised pre-training objective for human activity recognition and explore its effectiveness in comparison to state-of-the-art unsupervised learning techniques.

project image

Location based payload imaging


Apoorva, J and Mohan, Brinda and Beedu, Apoorva and Nayak, Mahendra M and Rao, Divya and Agrawal, VK
2015 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT)
pdf / bibtex /

We present a method to add the capability of imaging at any commanded latitude and longitude by including a provision to estimate the time required to reach the desired latitude and longitude using Location-Based Payload Imaging.


Teaching


Service

I have reviewed for BMVC(2021-2024), PerDream2023, VTTA2022.

Mentoring


Design and source code from Jon Barron's website