Viraj Prabhu

I am a fourth year CS Ph.D. student at Georgia Tech, advised by Judy Hoffman. My research interests are in developing data-efficient and resilient computer vision systems that can be deployed in the real world. Specifically, I am interested in label-efficient learning (particularly few-shot and active learning), adaptation across visual tasks and domains, and reliable and calibrated uncertainty estimation from deep neural networks.

I earned my Master's in CS (awarded the MS Research award) in Spring '19, also at Georgia Tech, where I was advised by Devi Parikh (and worked closely with Dhruv Batra) on developing visual conversational agents.

In grad school, I've had the opportunity to work on domain adaptation research at NVIDIA (with Sanja Fidler) and Salesforce (with Nikhil Naik). I also spent two wonderful summers doing research in machine learning for healthcare at Curai (with Anitha Kannan). Prior to joining Georgia Tech, I was a research assistant in the Machine Learning and Perception lab at Virginia Tech, advised by Dhruv Batra. Before that, I worked as a software developer at Adobe.

I received my Bachelor's degree in Computer Science from BITS Pilani. Over the course of my undergrad, I was fortunate to undertake research internships at Adobe, Tonbo Imaging, and CEERI Pilani, where I worked on problems ranging from video segmentation to camera calibration.

On the side, I have been an open-source contributor to CloudCV and served as mentor for Google Summer of Code (2016, 2017) and Google Code-In (2017) for the Fabrik project. I've also won a couple of hackathons (VTHacks '17, BITS Google Hackathon '14). Apart from work, I enjoy running, soccer, reading, playing the guitar, and (occasionally) writing.

Contact: I'm always happy to discuss research or grad school applications/life! Feel free to email me at virajp@gatech.edu.

Fall '11-Spring '15

Fall '17-current

Summer '14, Fall '15- Summer '16

Summer '18, '19

Summer '21

Summer '22

News

[May '23] Preprints on Language-guided Counterfactual Image Generation and adapting object detectors out on arXiv!.
[Jan '23] Gave an invited talk at Google Research Zurich on Reliable Computer Vision
[Sep '22] Adapting Self-supervised Vision Transformers was accepted at NeurIPS 2022!
[Jun '22] Gave a tutorial on Human-Centered AI for Computer Vision at CVPR 2022.
[May '22] Co-organizing the Learning from Limited and Imperfect Data workshop at ECCV 2022.
[Apr '22] Can domain adaptation make object recognition work for everyone? was accepted to L3D-IVU at CVPR '22.
[Oct '21] Papers on bias discovery and mitigation accepted at BMVC 2021!
[Oct '21] Recognized as an outstanding reviewer at NeurIPS 2021.
[Jul '21] Two papers on Unsupervised DA and Active DA accepted to ICCV 2021!

[Jul '21] Preprint on source-free domain adaptative semantic segmentation is out on arXiv.
[Jun '21] Recognized as an outstanding reviewer for CVPR 2021.
[Jan '21] Head TA for Intro to Computer Vision, Spring 2021.
[Dec '20] Preprint on Selective Entropy Optimization via Committee Consistency for Unsupervised DA is out on arXiv.
[Oct '20] Preprint on Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings is out on arXiv.
[Oct '19] Open Set Medical Diagnosis was accepted to the ML4H workshop at NeurIPS '19.
[Sep '19] Fabrik: An Online Collaborative Neural Network Editor will appear at the Workshop on AI Systems, SOSP '19.
[May '19] Completed my Master's degree!
[May '19] Few-Shot Learning for Dermatological Disease Diagnosis was accepted as a spotlight to MLHC 2019.
[Mar '19] Awarded the Georgia Tech College of Computing's MS Research award (1 student annually).
[Aug '18] Do Explanations make VQA Models more Predictable to a Human? was accepted to EMNLP 2018.
Reviewer for ICLR, CVPR, ECCV, NeurIPS (adjudged top-30% of reviewers) 2018.
[Apr '18] Just released our PyTorch implementation of Learning Cooperative Visual Dialog Agents with Deep Reinforcement Learning on GitHub.
[Jul '17] Presented a demo on Visual Chatbots at CVPR 2017.
[Jul '17] Presented It Takes Two to Tango: Towards Theory of AI's Mind at the Chalearn Looking at People Workshop at CVPR 2017.
[Jun '17] The Promise of Premise: Harnessing Question Premises in Visual Question Answering was accepted at EMNLP 2017.
[Jun '17] Evaluating Visual Conversational Agents via Cooperative Human-AI Games was accepted at HCOMP 2017.

Read Less

Publications

	LANCE: Stress-testing Visual Models by Generating Language-guided Counterfactual Images Viraj Prabhu, Sriram Yenamandra, Prithvijit Chattopadhyay, Judy Hoffman Paper Project Page
	Bridging the Sim2Real gap with CARE: Supervised Detection Adaptation with Conditional Alignment and Reweighting Viraj Prabhu, David Acuna, Andrew Liao, Rafid Mahmood, Marc T. Law, Judy Hoffman, Sanja Fidler, James Lucas Paper
	Adapting Self-Supervised Vision Transformers by Probing Attention-Conditioned Masking Consistency NeurIPS 2022 Viraj Prabhu, Sriram Yenamandra, Aaditya Singh, Judy Hoffman (* = equal contribution) Paper News
	Can domain adaptation make object recognition work for everyone? L3D-IVU workshop at CVPR 2022 Viraj Prabhu, Ramprasaath R. Selvaraju, Judy Hoffman, Nikhil Naik Paper
	AUGCO: Augmentation Consistency-guided Self-training for Source-free Domain Adaptive Semantic Segmentation Computer Vision in the Wild workshop, ECCV 2022 (spotlight) Viraj Prabhu, Shivam Khare, Deeksha Kartik, Judy Hoffman (* = equal contribution) Paper Video
	UDIS: Unsupervised Discovery of Bias in Deep Visual Recognition Models BMVC 2021 Arvind Krishnakumar, Viraj Prabhu, Sruthi Sudhakar, Judy Hoffman Paper Code
	Mitigating Bias in Visual Transformers via Targeted Alignment BMVC 2021 Sruthi Sudhakar, Viraj Prabhu, Arvind Krishnakumar, Judy Hoffman Paper
	Selective Entropy Optimization via Committee Consistency for Unsupervised Domain Adaptation ICCV 2021 Viraj Prabhu, Shivam Khare, Deeksha Kartik, Judy Hoffman Paper Project Page Code Video Slides Poster
	Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings ICCV 2021 Viraj Prabhu, Arjun Chandrasekaran, Kate Saenko, Judy Hoffman Paper Project Page Code Video Slides Poster
	Open Set Medical Diagnosis ML4H workshop at NeurIPS 2019 Viraj Prabhu, Anitha Kannan, Geoffrey J. Tso, Namit Katariya, Manish Chablani, David Sontag, Xavier Amatriain Paper
	Few-shot Learning for Dermatological Disease Diagnosis MLHC 2019 (spotlight), ML4H workshop at NeurIPS 2018 Viraj Prabhu, Anitha Kannan, Murali Ravuri, Manish Chablani, David Sontag, Xavier Amatriain Paper
	Do Explanations make VQA Models more Predictable to a Human? EMNLP 2018, Chalearn Looking at People Workshop, CVPR 2017 Arjun Chandrasekaran, Viraj Prabhu, Deshraj Yadav, Prithvijit Chattopadhyay, Devi Parikh (* = equal contribution) Paper
	The Promise of Premise: Harnessing Question Premises in Visual Question Answering EMNLP 2017 Aroma Mahendru, Viraj Prabhu, Akrit Mohapatra, Dhruv Batra, Stefan Lee ( equal contribution) Paper Code Dataset
	Evaluating Visual Conversational Agents via Cooperative Human-AI Games HCOMP 2017 Prithvijit Chattopadhyay, Deshraj Yadav, Viraj Prabhu, Arjun Chandrasekaran, Abhishek Das, Stefan Lee, Dhruv Batra, Devi Parikh Paper Code

Projects

Fabrik: An Online Collaborative Neural Network Editor Utsav Garg, Viraj Prabhu, Deshraj Yadav, Ram Ramrakhya, Harsh Agrawal, Dhruv Batra Lead mentor on Fabrik, an open-source web platform to collaboratively build, visualize, and design neural networks in the browser. Report Code
PyTorch implementation of Learning Cooperative Visual Dialog Agents with Deep Reinforcement Learning Nirbhay Modhe, Viraj Prabhu, Michael Cogswell, Satwik Kottur, Abhishek Das, Stefan Lee, Devi Parikh, Dhruv Batra Code
Adobe Captivate Prime During my time as a software developer at Adobe (Aug '15-'16), I was responsible for the Captivate Prime Android app through two release cycles. I developed features for offline content play-back, syncing, and UI.
Automated camera calibration and boresighting Over a research internship at Tonbo Imaging (Spring '15), I designed an algorithm for automated camera calibration and implemented a boresighting algorithm for the company's video precision boresight tool.
KeyframeCut Over an internship at Adobe, I co-developed KeyframeCut, a fast graphcut-based segmentation algorithm for real-time background substitution in video. Transferred into Magic Green Screen, the marquee feature of Adobe Presenter Video Express 11. Demo Blog

Fall '11-Spring '15

Fall '17-current

Summer '14, Fall '15- Summer '16

Summer '18, '19

Summer '21

Summer '22

News

Read More

Read Less

Publications

LANCE: Stress-testing Visual Models by Generating Language-guided Counterfactual Images

Bridging the Sim2Real gap with CARE: Supervised Detection Adaptation with Conditional Alignment and Reweighting

Adapting Self-Supervised Vision Transformers by Probing Attention-Conditioned Masking Consistency

Can domain adaptation make object recognition work for everyone?

AUGCO: Augmentation Consistency-guided Self-training for Source-free Domain Adaptive Semantic Segmentation

UDIS: Unsupervised Discovery of Bias in Deep Visual Recognition Models

Mitigating Bias in Visual Transformers via Targeted Alignment

Selective Entropy Optimization via Committee Consistency for Unsupervised Domain Adaptation

Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings

Open Set Medical Diagnosis

Few-shot Learning for Dermatological Disease Diagnosis

Do Explanations make VQA Models more Predictable to a Human?

The Promise of Premise: Harnessing Question Premises in Visual Question Answering

Evaluating Visual Conversational Agents via Cooperative Human-AI Games

Projects

Fabrik: An Online Collaborative Neural Network Editor

PyTorch implementation of Learning Cooperative Visual Dialog Agents with Deep Reinforcement Learning

Adobe Captivate Prime

Automated camera calibration and boresighting

KeyframeCut