|Title: PhD scholarship in object detection INRIA LEAR Team||Posted: May 16, 2014|
|Location: Grenoble, France|
|Department: LEAR team |
Description: The goal of this internship/PhD is to improve the current performance of object detection systems on several fronts.
In order to improve current object detectors, the objectives here are two-fold: (a) to make the representation more robust and distinctive and (b) to collect additional training data to better model the variability of object categories.
To improve the representation, which is currently either based on HOG descriptors, or local features like SIFT encoded by bag-of-word histograms or Fisher vectors, we will design mid-level or high-level features, for example based on image segmentation and contour features.
To collect additional training data, object detectors will be trained in a weakly supervised scenario, which will result in significantly more data for training. Initially, we will train object detectors from images for which no ground-truth object locations are known, but only a image-wide label that indicates the presence of one or more instances of the category in the image. We will, then, move to mixed scenarios where a few annotations are available.
Duration: 3-4 years
- Master degree (preferably in Computer Science or Applied Mathematics; Electrical Engineering will also be considered)
- Solid programming skills; the project involves programming in C
- Solid mathematics knowledge (especially linear algebra and statistics)
- Creative and highly motivated
- Fluent in English, both written and spoken
- Prior knowledge in the areas of computer vision, machine learning or data mining is a plus
Some recent papers that are relevant to this topic
Efficient Action Localization with Approximately Normalized Fisher Vectors, Oneata, Verbeek, Schmid, CVPR'14
- Multi-fold MIL Training for Weakly Supervised Object Localization, Cinbis, Verbeek, Schmid, CVPR'14
Segmentation Driven Object Detection with Fisher Vectors,
Cinbis, Verbeek, Schmid, ICCV'13
- Image Classification with the Fisher Vector: Theory and Practice, Sanchez, Perronnin, Mensink, Verbeek, IJCV'13
- Regionlets for Generic Object Detection, Wang, Yang, Zhu, Lin, ICCV'13
Segment, Classify and Search Objects Locally, Li, Gavves, van de Sande, Snoek, Smeulders, ICCV'13
- Bottom-up Segmentation for Top-down Detection, Fidler, Mottaghi, Yuille, Urtasun, CVPR'13
Application Instructions: Please send applications via email to
- Jakob Verbeek (Jakob.Verbeek@inria.fr) and
- Cordelia Schmid (Cordelia.Schmid@inria.fr)
- a complete CV
- grades during for MSc courses and thesis
- two letters of reference