Tarun Kalluri

I am a PhD candidate in the CSE department at UCSD advised by Prof. Manmohan Chandrakar. My PhD research focuses on learning label-efficient representations from large-scale images and videos that are useful and transferable across a wide range of domains.

Recently, my focus is on studying the robustness and fairness properties of Generative AI and foundation models. I am fortunate to have interned at Google Research and Facebook AI Research during my PhD study. I also recieved the IPE PhD fellowship in 2021.

Tarun Kalluri

CV / Research Statement
Email / Google Scholar / GitHub / LinkedIn

News

July 2024: One paper on domain adaptation accepted to ECCV 2024 in Milan, Italy.
May 2024: One paper on language-guided domain transfer accepted to ICML 2024 in Vienna.
April 2024: Two papers on open-world instance segmentation and synthetic data for robust disaster assessment accepted to workshops in CVPR.
Jan 2024: Successfully qualified the PhD candidacy exam, and actively looking for full-time positions.
June 2023: Started summer internship at Google Research in Mountain View campus!
May 2023: We are conducting a workshop on geographical robustness in computer vision at ICCV 2023, please see the website for details!
Mar 2023: GeoNet accepted to CVPR 2023!
Jan 2023: FLAVR got selected as the Best Paper Finalist WACV 2023!
More
Dec 2022: Selected as Highlighted reviewer at ICLR2022 and Neurips 2022.
July 2022: MemSAC accepted to ECCV 2022.
June 2021: Started internship at FAIR, MPK to work on open-world learning.
May 2021: One paper on domain adaptation accepted at CVPR 2021!
Sep 2020: Recipient of IPE PhD fellowship for the year 2020-21.
Aug 2020: Our paper on training neural network for molecular force predictions is accepted to the Journal of Physical Chemistry.
June 2020: Joined Facebook AI as a summer intern with the multimodal learning group.
Dec 2019: One paper on datasets for resource constrained semantic segmentation accepted to NCVPRIPG 2019.
Sep 2019: Moved to sunny San Diego! Starting as a PhD student at CSE department in UCSD from Fall 2019. I will be working with the Visual Computing Group.
Jul 2019: Our paper on Universal Semantic Segmentation is accepted to ICCV 2019.

Selected Publications (Google Scholar for the full list)
UDA-Bench: Revisiting Common Assumptions in Unsupervised Domain Adaptation Using a Standardized Framework
Tarun Kalluri , Sreyas Ravichandran, Manmohan Chandraker.
ECCV, 2024
arxiv / pdf / Code / reviews / poster

New standardized training framework for unsupervised adaptation with several practical insights for improving performance of domain adaptation methods.

Tell, Don`t Show!: Language Guidance Eases Transfer Across Domains in Images and Videos
Tarun Kalluri , Bodhisattwa Prasad Majumder, Manmohan Chandraker.
ICML, 2024
webpage / arxiv / pdf / Code / reviews / poster

Language guided transfer using natural text descriptions to improve unlabeled domain robustness in images and videos.

Robust Disaster Assessment from Aerial Imagery Using Text-to-Image Synthetic Data
Tarun Kalluri , Jihyeon Lee, Kihyuk Sohn, Sahil Singla, Manmohan Chandraker, Joseph Xu, Jeremiah Liu
Generative Models Workshop, CVPR, 2024
arxiv / pdf/ poster

Synthetic data generation using text-to-image models for robust disaster assessment across geographies from aerial imagery using limited labeled data.

GeoNet: Benchmarking Unsupervised Adaptation Across Geographies
Tarun Kalluri , Wangdong Xu, Manmohan Chandraker.
CVPR, 2023
webpage / arxiv / pdf / code / reviews / poster/ dataset

New large-scale dataset to study generalization ability of vision models across diverse geographies.

Open-world Instance Segmentation: Top-down Learning with Bottom-up Supervision
Tarun Kalluri , Weiyao Wang, Heng Wang, Manmohan Chandraker, Lorenzo Torresani, Du Tran
L3D-IVU Workshop, CVPR, 2024
project page / arxiv / pdf / code / poster

Open-world instance segmentation using a top-down framework assisted by bottom-up supervision.

MemSAC: Memory Augmented Sample Consistency for Large Scale Domain Adaptation
Tarun Kalluri , Astuti Sharma, Manmohan Chandraker.
ECCV, 2022
webpage / arxiv / pdf / code / reviews / poster

Memory-based consistency losses to scale unsupervised domain adaptation to a large number of classes, including fine-grained datasets.

FLAVR: Flow-Agnostic Video Representations for Fast Frame Interpolation

Selected as finalist for Best Paper Award
(Top 12 out of 641 papers, Top 2%).

Tarun Kalluri , Deepak Pathak, Manmohan Chandraker, Du Tran
WACV, 2023
MVA Special Issue on WACV Award Finalists, 2023
arxiv / MVA SI Award Edition / pdf / project webpage / video / Code/ reviews / Colab

Fast, accurate and flow-free video frame interpolation technique using space-time convolutions, capable of single shot multi frame prediction.

Instance Level Affinity Based Transfer for Unsupervised Domain Adaptation
Astuti Sharma, Tarun Kalluri , Manmohan Chandraker.
CVPR, 2021
arxiv / poster / code

Proposes multi sample contrastive loss using instance level similarities across source and target domains for robust feature alignment and knowledge transfer.

Universal Semi-Supervised Semantic Segmentation
Tarun Kalluri , Girish Varma, Manmohan Chandraker, CV Jawahar
ICCV, 2019
pdf / poster / reviews / code

Addresses geographical disparities between semantic segmentation datasets with minimum labeling overhead from each domain.


Template modified from Jon Barron and Micheal Hahn