 |
HYPERION (2012-2014) Large-scale statistical learning for visual recognition |
Principal Investigators :
- Dr. Zaid Harchaoui, LE AR project-team, Inria Grenoble Rhône Alpes
- Dr. Cordelia Schmid, LEAR project-team, Inria Grenoble Rhône Alpes
- Prof. Jitendra Malik, University of California Berkeley
- Prof. Noureddine El Karoui, University of California Berkeley
Research objectives:
A recent trend in computer vision as well as in other fields is the advent of “big data“ with an increasing number of large annotated image and video datasets now becoming available to researchers, along with an increasing number of benchmarks and challenges associated to these datasets. In particular, designing principled and scalable statistical learning approaches from such big datasets is actually challenging. Furthermore, these datasets also bring to light not only major statistical and computational challenges but also core computer vision issues. The scientific objective of HYPERION is to take up these challenges and propose new principled large-scale statistical learning approaches for image interpretation and learning models for video understanding.
Scientific achievements:
- Action localization with movemes: The team designed new visual descriptors, called movemes, blending information from shape and motion cues, for human action localization.
- Fast and robust archetypal analysis: The team proposed a novel approach, called robust archetypal analysis, for visualization of large image collections.
- Object localization with S-LSVM: The team proposed a two-step approach called S-LSVM for object localization in images with minimal supervision.
Publications and Awards:
- 10 Conference papers.
- Open-source software packages and publicly available benchmark datasets.
- Awardee of a 2012 France-Berkeley Fund.
Selected publication:
Y. Chen, J. Mairal, and Z. Harchaoui. Fast and robust archetypal analysis for representation learning. In CVPR Computer Vision and Pattern Recognition, 2014.
Follow up :
The collaboration has been fruitful, leading to several published papers in top venues in computer vision and machine learning. Follow-up work is thus anticipated along the research themes that were investigated as part of HYPERION.
