Nowadays, the computer vision community relies on deep learning algorithms that have been trained to perform various tasks and learned a the necessary concepts on large amounts of data. However, if one needs to target new tasks, it is necessary to redevelop an appropriate dataset which is costly as the annotation process is expensive. We propose to the prospective intern to work on a deep learning algorithm that would improve the state of the art in guiding the annotation process in the field of pedestrian studies.
The work is performed at Paris Sud University in the context of the ANR MOHICANS research project.
The work will be performed on a crowd data set as illustrated in the attached document. Our team finished 4th on the CVPR 2019 MOT challenge. We also have a new dataset, and we need, just as the whole community, better and more streamlined algorithms for interactive annotation. The field is very competitive and the internship work is potentially publishable in top conferences. Access to adequate computing resources for carrying out the research will be provided.
For more details: read the following document
Contact and application: send your application (CV, cover letter and academic transcripts) to firstname.lastname@example.org and email@example.com
(c) GdR 720 ISIS - CNRS - 2011-2020.