Personal tools
Laboratoire d'InfoRmatique en Image et Systèmes d'information

Skip to content. | Skip to navigation

Laboratoire d'InfoRmatique en Image et Systèmes d'information
UMR 5205 CNRS / INSA de Lyon / Université Claude Bernard Lyon 1 / Université Lumière Lyon 2 / École Centrale de Lyon
You are here: Home > News > LIRIS News > Natalia Neverova is the winner of the thesis prize for Signal, Image and Vision of Club EEA, GRETSI and GdR ISIS

Natalia Neverova is the winner of the thesis prize for Signal, Image and Vision of Club EEA, GRETSI and GdR ISIS

Natalia Neverova is the winner of the thesis prize for Signal, Image and Vision of Club EEA, GRETSI and GdR ISIS for the year 2017 for her thesis at LIRIS and INSA-Lyon (Oct. 2012 - Apr. 2016) named "Deep Learning for Human Motion Analysis ».

This thesis was  
- supervised by Christian Wolf, Maître de Conférences, HDR, à l'université de Lyon/INSA Lyon,
- and co-supervised by Graham W. Taylor, assistant Professor, University of Guelph, Canada.

Natalia Neverova started her PhD in October 2012 at LIRIS' Imagine research group. The objective of the thesis, funded by a project of type « Investissement’s d'Avenir »  and lead by French robotics company Awabot, was to equip a mobile robot with gesture interaction capabilities in order to bring an emotional dimension into the project.

Natalia has worked on Deep Learning for human motion, with several applications:
- Gesture recognition
- Estimation of articulated pose
- Authentication of smartphone users (in association with Google USA).

She proposed several models and innovative training algorithms:

- Her work on gesture recognition  won two scientific competitions associated with the ECCV 2014 and CVPR 2015 conferences. Miss Neverova also demonstrated the theoretical properties of the learning algorithm.

- For the estimation of articulated pose, Ms Neverova proposed an algorithm that is both semi-supervised and weakly supervised, allowing training from heterogeneous data. A great effort was made in a practical and experimental way to obtain, from simulations, a database of 70 000 densely annotated synthetic images.

- The problem of biometrics was tackled by an approach merging signal processing with machine learning. The proposed model allows a multi-scale and invariant decomposition of the signal, while being completely trainable.

During her thesis, Ms. Neverova did two research internships abroad:
- as part of a collaboration with Graham W. Taylor, co-supervisor of the thesis, a 3-month stay at the University of Guelph, Canada.
- a 3-month research trip to Google, Mountain View.

She is currently pursuing her career in the Parisian team of Facebook AI Research.


nullnull

pour en savoir plus :

Document Actions