Tutorial at WACV 2025
AZ Ballroom Salon 4
28th Feb 2025, 13:00-17:00 (MST | GMT-7)
Ashish Tiwari and Shanmuganathan Raman
Computer Vision, Imaging, and Graphics Lab, IIT Gandhinagar
Abstract
Passive depth sensing in computer vision is gaining traction due to advantages over active sensors, such as lower cost, reduced power consumption, enhanced portability, and greater depth range. However, these benefits come with increased computational complexity. Traditional geometric methods like multi-view stereo and structure from motion achieve high-quality 3D reconstructions but often struggle with capturing surface and sub-surface reflectance properties, leading to incomplete reconstructions.
Photometric methods, including Photometric Stereo and Shape from Polarization, offer a solution by enabling finer geometric detail recovery and the joint estimation of shape, material, and reflectance properties. These techniques extend beyond opaque surfaces to include transparent, translucent, and textureless objects. They also facilitate more complete 3D reconstructions by leveraging changes in light direction and density due to refraction within objects—an aspect often overlooked by traditional geometric approaches. Understanding how refraction modifies image acquisition geometry enhances the depth and accuracy of 3D scene analysis, presenting a rich area for future research.
This tutorial aims to provide the computer vision and graphics community with fresh perspectives on reconstructing the geometry, appearance, and reflectance of complex scenes through photometric inverse rendering. It will offer practical insights for applying these methods in real-world scenarios, paving the way for advancements across various applications such as physics-based rendering, machine vision, and virtual reality. By addressing these challenges, the tutorial extensively explores techniques that expand the possibilities in 3D shape recovery.
Organization
Speakers
Ashish is a PhD student in Electrical Engineering, being advised by Dr. Shanmuganathan Raman at IIT Gandhinagar. His research interest lies in photometric inverse rendering through methods such as Photometric Stereo, Shape from Polarization, Photo-polarimetric Stereo, and Reconstruction from Shadows. He is the recipient of the Prime Minister's Research Fellowship (PMRF) and the Qualcomm Innovation Fellowship (QIF) for 2023-24. Currently, he is on the Fulbright Nehru Doctoral Research (FNDR) Fellowship 2024-25 and is working with the Computational Imaging Group at Rice University. He has published his research on photometric inverse rendering across popular venues such as ECCV, SIGGRAPH Asia, WACV, ICASSP, and ICPR, with the best poster award at Pacific Graphics 2023.
Shanmuganathan Raman is the Jibaben Patel Chair (AI) Professor in EE and CSE at IIT Gandhinagar. He obtained a BE degree from PSG Tech, MTech, and PhD from IIT Bombay. His research interests are Computer Vision and Computer Graphics. He obtained a Microsoft Research PhD Fellowship in 2007, an Excellence in PhD Thesis Work Award at IIT Bombay in 2012, an Excellence in Research Work Award from IIT Gandhinagar in 2022, and a Qualcomm Innovation Fellowship in 2023. He has published more than 100 peer-reviewed articles in journals and conferences of repute in Computer Vision and Computer Graphics. He is a recipient of several sponsored research projects including SERB Core Research Grant, SERB IMPRINT-2, ISRO RESPOND, and SERB MATRICS. He was the course coordinator for 3D Digitization for Cultural Heritage at IIT Gandhinagar, which inaugurated the GIAN program across India.