Anand Bhattad

Research Assistant Professor
Toyota Technological Institute at Chicago

Email / CV / Google Scholar / Twitter

I am a Research Assistant Professor at the Toyota Technological Institute at Chicago (TTIC) located in the University of Chicago campus. Earlier, I completed my PhD from Computer Vision Group at University of Illinois Urbana-Champaign (UIUC) with my advisor David Forsyth. During my PhD, I had the pleasure of closely collaborating with Derek Hoiem, Shenlong Wang, and Yuxiong Wang at UIUC.

News

Sept 2023: Two papers accepted by NeurIPS 2023. One on discovering hidden gems in StyleGAN and another on language guided 3D aware image editing.

Sept 2023:

July 2023: One paper accepted by ICCV 2023 on Equivariant Dense Prediction Models.
Jun 2023: Successfully organized Scholars and Big Models workshop at CVPR 2023.
May 2023: Defended my PhD!

Recent and Upcoming Talks

Stanford University, Jun 2023
University of Tübingen, Autonomous Vision Group, May 2023
UC Berkeley: Vision Seminar, Apr 2023
NVIDIA Research, Apr 2023
MIT: Vision and Graphics Seminar, Apr 2023
CMU: VASC Seminar, Mar 2023
UW: Vision Seminar, Mar 2023
UMD: Vision Seminar, Mar 2023
UCSD: Pixel Cafe Seminar, Feb 2023
TTIC: Research Talk, Feb 2023

Research

My research interests are in Computer vision, graphics, photography and machine learning, with a focus on understanding the fundamentals of scene representations and generative models. My research aims to uncover the emergent scene properties in generative models and establish them as the backbone for modern computer vision systems. I am passionate about developing data-driven methodologies for understanding, modeling, and recreating the visual world.

Publications

Preprints / Technical Reports

	UrbanIR: Large-Scale Urban Scene Inverse Rendering from a Single Video Zhi-Hao Lin, Bohan Liu, Yi-Ting Chen, David Forsyth, Jia-Bin Huang, Anand Bhattad, Shenlong Wang arXiv, 2023 [arXiv] [project page] UrbanIR creates realistic 3D renderings of urban scenes from single videos, allowing for novel lighting conditions and controllable editing.
	Blocks2World: Controlling Realistic Scenes with Editable Primitives Vaibhav Vavilala, Seemandhar Jain, Rahul Vasanth, Anand Bhattad, David Forsyth arXiv, 2023 arXiv Blocks2World decomposes 3D scenes into editable primitives and uses a trained model to render these into 2D images, providing high control for scene editing.
	StyLitGAN: Prompting StyleGAN to Produce New Illumination Conditions Anand Bhattad, David A. Forsyth arXiv, 2023 [arXiv] [project page] By imposing known physical facts about images, we can prompt StyleGAN to generate relighted or resurfaced images without using labeled data.
	Make It So: Steering StyleGAN for Any Image Inversion and Editing Anand Bhattad, Viraj Shah Derek Hoiem, David A. Forsyth arXiv, 2023 [arXiv], [project page] A novel near-perfect GAN Inversion method that preserves editing capabilities, even for out-of-domain images
	Illuminator: Learning to Correct Lighting without Ground Truth Data Anand Bhattad, Brian Chen, Shenlong Wang, David A. Forsyth First self-supervised, image-based object relighting method trained without labeled paired data, CGI data, geometry, or environment maps.
	SIRfyN: Single Image Relighting from your Neighbors David A. Forsyth, Anand Bhattad, Pranav Asthana, Yuani Zhong, Yuxiong Wang arXiv, 2022 Technical Report First scene relighting method that requires no labeled or paired image data.

	StyleGAN knows Normal, Depth, Albedo and more Anand Bhattad, Daniel McKee, Derek Hoiem, David Forsyth NeurIPS, 2023 [arXiv] StyleGAN has easy accssible internal encoding of intrinsic images as originally defined by Barrow and Tenenbaum in their influential paper of 1978.
	OBJect 3DIT: Language-guided 3D-aware Image Editing Oscar Michel, Anand Bhattad, Eli VanderBilt, Ranjay Krishna, Ani Kembhavi, Tanmay Gupta NeurIPS, 2023 [arXiv] [project page] A synthetic dataset and a model that learns to rotate, translate, insert, and remove objects identified by language in a scene. It can transfer to real-world images.
	Improving Equivariance in State-of-the-Art Supervised Depth and Normal Predictors Yuani Zhong, Anand Bhattad, Yuxiong Wang David A. Forsyth ICCV 2023 SOTA normal and depth predictors are not equivariant to image cropping. We propose equivariant regularization loss to improve equivariance in these models.
	Cut-and-Paste Object Insertion by Enabling Deep Image Prior for Reshading Anand Bhattad, David A. Forsyth 3DV, 2022 [project page] Convincing cut-and-paste reshading with consistent image decomposition inferences.
	DIVeR: Real-time and Accurate Neural Radiance Fields with Deterministic Integration for Volume Rendering Liwen Wu, Jae Yong Lee, Anand Bhattad, Yuxiong Wang, David A. Forsyth CVPR, 2022 (Best Paper Finalist) [project page] / Training Code / Real-time Code Improving Real-Time NeRF with Deterministic Integration.
	View Generalization for Single Image Textured 3D Models Anand Bhattad, Aysegul Dundar, Guilin Liu, Andrew Tao, Bryan Catanzaro CVPR, 2021 [project page] Consistent textured 3D inferences from a single 2D image.
	Unrestricted Adversarial Examples via Semantic Manipulation Anand Bhattad, Min Jin Chong, Kaizhao Liang, Bo Li, David A. Forsyth ICLR, 2020 CVPR-W on Adversarial ML in Real-World Computer Vision Systems, 2019 Generating realistic adversarial examples by image re-colorization and texture transfer.
	Improved Style Transfer with Calibrated Metrics Mao Chuang Yeh, Shuai Tang, Anand Bhattad, Chuhang Zou, David A. Forsyth WACV, 2020 A novel quantitative evaluation procedure for style transfer methods.
	Detecting Anomalous Faces with "No Peeking'' Autoencoders Anand Bhattad, Jason Rock, David A. Forsyth CVPR Workshop on Vision with Biased or Scarce Data, 2018 A simple unsupervised method for detecting anomalous faces by carefully constructing features from "No Peeking" or inpainting autoencoders.

Teaching

Graduate Teaching Assistant, CS498 Applied Machine Learning, Fall 2018

Graduate Teaching Assistant, CS225 Data Structures, Spring 2017

Graduate Teaching Assistant, CS101 Intro Computer Science, Spring 2016 (Ranked as Outstanding TA) & Fall 2017

Template credit: Jon Barron.