On Tuesday 16th June, Jean Ponce and Yasutaka Furukawa were awarded the Longuet-Higgins prize for their 2007 article at the IEEE Conference on Computer Vision and Pattern Recognition (watch a video of the awards ceremony).
The Longuet-Higgins prize
The Longuet-Higgins prize is awarded each year at the Conference on Computer Vision and Pattern Recognition (CVPR) in recognition of one or two articles which have had a significant impact on the scientific community over the course of the last 10 years (after having been published at this same conference ten years earlier). The prize is awarded to fundamental contributions made to computer vision which have successfully passed “the test of time”.
The article was about an algorithm that was patented in the USA in 2012 (“Match, expand and filter technique for multi-view stereopsis”). The PMVS (Patch-Based Multi-View Stereo) program, which employs the algorithm, works by capturing a set of camera parameters and images before rebuilding the 3D structure of an object or a scene that is visible on these images.
“PMVS takes images and the calibration parameters from the cameras used to take them”, explains Jean Ponce. “It then compares small sections of these images (the “patches” from the acronym) in order to build robust matching hypotheses, which take into account both local constraints (photo-coherence) and global visibility constraints. This has been the key to the success of PMVS, which builds the scene that was observed based on these hypotheses.”
The program has been integrated into Photo Tours/Google Maps, and has been used in the special effects industry by Industrial Light & Magic and Weta Digital. It has also been used by Cicero Moraes in Brazil for facial reconstruction.
PMVS has been one of my team’s biggest successes. It is widely used in industry and in academia, and its use has led to collaborations with Hélène Dessales from the archaeology department at the ENS and with Yves Ubelmann from the company Iconem in the preservation of cultural heritage, which I have found highly rewarding.
The company Maxon took out a commercial licence for PMVS, which is available here for non-commercial uses.
It's a real honour to be awarded this prize, and I am really happy that it also recognises Yasutaka, who is now a teacher-researcher at Simon Fraser University in Vancouver (Canada) and whose outstanding PhD research led to the development of PMVS.
A renowned researcher
Jean Ponce is a researcher in the field of artificial vision and head of the joint project team Willow. (CNRS, ENS, Inria). He is also the scientific director of the Prairie Institute in Artificial Intelligence.
Willow, a shining example of a project team
The research carried out by Willow focuses on issues linked to representation in the field of visual recognition. The team's researchers have been awarded prizes at the three main international conferences on vision (CVPR, ECCV, ICCV) and one of the two main machine learning conferences (ICML). The team’s reputation is growing, having been awarded a total of 10 different prizes.
Find out more:
Ivan Laptev (1): ICCV Helmholtz prize, 2017
Jean Ponce (3): CVPR Longuet-Higgins prize (with Cordelia Schmid), 2016; ICML test-of-time award (with Francis Bach and Julien Mairal), 2019; CVPR Longuet-Higgins prize, 2020.
Josef Sivic (3): CVPR Longuet-Higgins prize, 2017; 2 ICCV Helmholtz prizes, 2017.
Cordelia Schmid (4): CVPR Longuet-Higgins prize 2006, 2014, 2016 (with Jean Ponce), ECCV Koenderink prize, 2018.
*A joint undertaking involving the CNRS, ENS and Inria