By Meredith Alexander Kunz, Adobe Research
Three Adobe Research scientists have been awarded the Helmholtz Prize by the Institute of Electrical and Electronics Engineers (IEEE) for research contributions that have had a significant impact on the field of computer vision. Adobe Research’s winners are research scientist Bryan Russell, principal scientist Eli Shechtman, and consulting senior research scientist Josef Sivic. The award was presented at the 2017 IEEE-sponsored International Conference on Computer Vision (ICCV) in Venice, Italy.
The prize recognizes research papers presented at the conference ten years ago or more that continue to be influential today. One such paper was Discovering Objects and Their Location in Images, published at ICCV 2005. Russell worked with Sivic and other contributors on this groundbreaking research. They were the first to pose the problem of unsupervised discovery of objects depicted in still images. The researchers asked, “can computers learn to identify objects simply by looking at images, even without labels or annotations?”
The team built on previous work in natural language processing—probablistic latent semantic analysis (pLSA) and a “bag-of-words” approach—to enable machines to find object categories. “The overall goal of unsupervised object discovery is still very relevant today as we seek to train recognition systems from the vast stores of available images without human annotations,” explains Russell.
Shechtman’s contribution, Actions as Space-Time Shapes, also from ICCV 2005, was among the earliest papers to address the problem of action recognition. The team of researchers represented human actions as three-dimensional shapes in space-time volume, a novel approach. They then created a mathematical formulation for understanding actions.
The team also collected a new action recognition dataset, one of the first in this domain, earning many citations from other researchers. “The dataset was perhaps the most important factor for advancing the field,” Shechtman says.
Sivic’s winning paper, Video Google: A Text Retrieval Approach to Object Matching in Videos, published at ICCV 2003, focused on searching videos for objects. The pair of researchers aimed to retrieve video frames containing a particular object given an image of the object as a query. They developed a visual indexing method adopting ideas from text retrieval, allowing a user to search for images in the same way that people use Google to retrieve websites containing specific words. They showed that objects could be effectively retrieved in a movie database using this approach, even when viewpoints on the object changed significantly between the query image and the movie frames in the database.
This work opened up the possibility of visual search for objects in much larger image and video collections than was possible before. Results of this research have been adopted by others in computer vision and related fields to enable, for example, searching for products using images taken by a mobile phone or for visual localization for autonomous vehicles.
About Adobe Research
With more than 120 world-class research scientists and engineers, Adobe Research blends cutting-edge academic discovery with industry impact. Our scientists are provided with the resources, support, and freedom to shape their ideas into innovative technologies. They collaborate with colleagues at over fifty universities, often presenting their work at international conferences. Many of our researchers’ discoveries are incorporated into Adobe’s products, building the company’s reputation as a pioneer in content and data intelligence. Today, Adobe Research is growing in new areas, including artificial intelligence, machine learning, immersive media, document intelligence, analytics, and data mining. Learn more at research.adobe.com.
About the Helmholtz Prize
Named after the nineteenth-century physician and physicist Hermann von Helmholtz, this prize is awarded every other year. It is presented at ICCV, a premier conference in the field of computer vision sponsored by the Institute of Electrical and Electronics Engineers (IEEE). Winners are selected by the IEEE’s Computer Society’s Technical Committee on Pattern Analysis and Machine Intelligence. The prize honors papers and their authors from conferences ten or more years ago that have advanced research in the field of computer vision.