Visual content is everywhere, video, photographs, media assets, scanned document images, and so on. Our vision research provides enabling technologies to give the gift of sight to automated processes. Combined with textual information from the images themselves (e.g. OCR, natural text) or external sources (web pages, internal database records) our research deals with all types of interaction where visual content is one component.
We have several lines of research including some of the areas of focus listed below. Our research combines skills in machine learning, pattern recognition and computer vision. Our research is often pursued in collaboration with external partners from industry, government and academia.
- Learning Visual Representations
- Visual Search
- Object Detection
- Video Analysis
- Visual Classification
- Subjective Visual Attributes
- We have 3 papers accepted at ICCV 2017.
- We won the MediaEval 2017 Retrieving Diverse Social Images Task challenge.
- We have 4 papers accepted at CVPR 2017.
- CVPR Outstanding Reviewer Award for Diane Larlus and Jérôme Revaud.
- Florent Perronnin is co-organizing, with Adrien Gaidon and Antonio Lopez, an IJCV special issue on “synthetic visual data”. This special issue follows the workshop on Virtual/Augmented Reality for Visual Artificial Intelligence (VARVAI) which was held in conjunction with ECCV 2016.
- ECCV Outstanding Reviewer Award for Diane Larlus
- We have 2 papers accepted at ECCV 2016 on action recognition (PDF) and instance-level image retrieval (PDF, webpage).
- We have 1 paper accepted at BMVC 2016 on object part transfer (PDF).
- We will co-organize a workshop on Virtual/Augmented Reality for Visual Artificial Intelligence at ECCV 2016.
- We have 2 papers accepted at CVPR 2016 on virtual worlds (PDF) and saliency estimation (PDF).
- We will co-organize a tutorial on "Scene-Text Localization, Recognition, and Understanding" at DAS 2016.
- ICCV Outstanding Reviewer Award for Naila Murray and Jerome Revaud.
- We have 1 paper accepted at ICCV 2015 on semantic word embeddings.
- We have 2 papers accepted at BMVC 2015 on extracting network gradient features and on multi-object tracking (oral).
- CVPR Outstanding Reviewer Award for Adrien Gaidon and Albert Gordo.
- We have 2 papers accepted at CVPR 2015 on hybrid networks and mid-level word image representations.
- ECCV 2014 Outstanding Reviewer Award for Florent Perronnin
- We had 3 papers accepted at CVPR 2014 on feature pooling, saliency prediction and training set augmentation. and 1 paper accepted at the workshop on Computer Vision and Human Computation.
- We co-organized a tutorial on Large-Scale Visual Recognition (LSVR) at CVPR 2014.
Other highlights in computer vision
Applications and Projects
Our research feeds into both current business process and blue-sky future disruptive technologies. We interface across the range of domains within Xerox from Smarter Document Management Technologies and processes, healthcare applications or the Transportation Solutions group as some examples. We also provide technologies for clients of Xerox who have specific needs for a visual based solution.
We participate heavily in open innovation (OI) projects, and have several current and past OI projects including PhD studentships and grant-based projects funded by ANR or the EU. As an example of the range of topics we research in both internal and OI projects, past work has included a navigation system for visual search based on implicit feedback from the eye-movements of a user; analysis, classification and organization of multilingual rich-content documents; handwritten word-spotting, hybrid classification of scanned and OCR'd documents, automatic image enhancement, and many more!
Some of the work of the group can be tried out on the imaging services part of Open Xerox. Directly connected to some current research streams are: