moteur de recherche

Atelier doctorants A3SI
Atelier doctorants A3SI
4-Feb-2021 14:30
Age: 25 days

L'atelier doctorant du jeudi 4/2 sera accessible en présentiel, à ESIEE Paris, et en distanciel. Si vous souhaitez participer à l'atelier en présentiel, il est nécessaire de vous inscrire en contactant Benjamin Perret au plus tard le mardi 2 février. En raison des restrictions dues à l'état sanitaire actuel, nous ne pourrons pas garantir l'accès à la salle aux personnes non inscrites. Le numéro de salle et le lien de connection seront donnés dans un prochain mail.

Le programme de l'atelier est le suivant:

  • 14h00 - 14h30: Yang Xiao, "3D Pose Estimation for Objects with Limited Annotations in the Wild "
  • 14h30 - 15h00: Thanh Nguyen, "Astronomical Object Detection with Morphology and Deep Learning"
  • 15h00 - 15h30 :Tom Monnier, "Deep Transformation-Invariant Clustering"
  • 15h30 - 16h00: Elias Barburo, "Towards Software Programmable Streaming Coarse Grained Reconfigurable Architectures efficient re-use"

Résumés des présentations:


Orateur : Yang Xiao, doctorant encadré par  Mathieu Aubry (ENPC) et Renaud Marlet (ENPC /

Titre de la présentation : 3D Pose Estimation for Objects with Limited Annotations in the Wild

Résumé : Detecting objects and estimating their viewpoints in images are key tasks of 3D scene understanding. Recent approaches have achieved excellent results on very large benchmarks for object detection and viewpoint estimation. However, performances are still lagging behind for novel object categories with few samples. In this paper, we tackle the problems of few-shot object detection and few-shot viewpoint estimation. We propose a meta-learning framework that can be applied to both tasks, possibly including 3D data. Our models improve the results on objects of novel classes by leveraging rich feature information originating from base classes with many samples. A simple joint feature embedding module is proposed to make the most of this feature sharing. Despite its simplicity, our method outperforms state-of-the-art methods by a large margin on a range of datasets, including PASCAL VOC and MS COCO for few-shot object detection, and Pascal3D+ and ObjectNet3D for few-shot viewpoint estimation. Moreover, for the first time, we tackle the combination of both few-shot tasks, on three challenging viewpoint estimation in the wild benchmarks, ObjectNet3D, Pascal3D+ and Pix3D, showing very promising results.


Orateur : Thanh Nguyen, doctorant encadré par Giovanni Chierchia (ESIEE Paris), Laurent Najman (ESIEE Paris), Benjamin Perret (ESIEE Paris) et Hugues Talbot (CentraleSupélec) 

Titre de la présentation : Astronomical Object Detection with Morphology and Deep Learning

Résumé : This work proposes a region-based convolutional neural network (R-CNN) approach to detect and segment astronomical objects. We then discuss a hybrid approach that takes the advantages of both morphological-based and machine learning-based models to adapt to astronomical contexts. The initial experiments suggest that the results of the proposed methods are significantly better than the results of the state-of-the-art methods on real and simulated astronomical images.  


Orateur :  Tom Monnier, doctorant encadré par Mathieu Aubry (ENPC) 

Titre de la présentation : Deep Transformation-Invariant Clustering

Résumé : Recent advances in image clustering typically focus on learning better deep representations. In contrast, we present an orthogonal approach that does not rely on abstract features but instead learns to predict image transformations and directly performs clustering in pixel space. This learning process naturally fits in the gradient-based training of K-means and Gaussian mixture model, without requiring any additional loss or hyper-parameters. It leads us to two new deep transformation-invariant clustering frameworks, which jointly learn prototypes and transformations. More specifically, we use deep learning modules that enable us to resolve invariance to spatial, color and morphological transformations. Our approach is conceptually simple and comes with several advantages, including the possibility to easily adapt the desired invariance to the task and a strong interpretability of both cluster centers and assignments to clusters. We demonstrate that our novel approach yields competitive and highly promising results on standard image clustering benchmarks. Finally, we showcase its robustness and the advantages of its improved interpretability by visualizing clustering results over real photograph collections.


Orateur :  Elias Barburo, doctorant encadré par Eva Dokladalova (ESIEE Paris) et Thierry Grandpierre (ESIEE Paris)  

Titre de la présentation : Towards Software Programmable Streaming Coarse Grained Reconfigurable Architectures efficient re-use

Résumé : Coarse-Grained Reconfigurable Architectures (CGRA) are designed to deliver high performance while drastically reducing the latency of the computing system. There are several types of CGRA according to the structure, application, type of resources, and memory infrastructure. We focus our work on a subset of CGRA designs that we call Software Programmable Streaming Coarse Grained Reconfigurable Architectures (SPS-CGRA). A SPS-CGRA is a more or less complex array of coarse-grained heterogeneous hardware resources with a coarser granularity than the classical. A SPS-CGRA can perform spatial and temporal computations at low latency. Its stream-based processing provides high performance maintaining a level of flexibility. Although they are often highly domain-specifically optimized, they keep several levels of custom post-fabrication programmability, given by a set of parameters, so that they can be reused. However, their re-use is generally limited due to the complexity of identifying the best allocation of the processing tasks into the hardware resources. Another limiting point is the complexity to produce a reliable performance analysis for each new implementation since no mature tool exist.

To solve these problems, we propose a complete mapping and scheduling framework that targets SPS-CGRA. We introduce a generic hardware model allowing one to express these intrinsically custom levels of flexibility, as well as data access and system configuration control, often neglected in existing works. We also propose a performance estimation analysis, based on the resource’s latency, for the upper bound of the computing cost. To complete, we present four different solutions for the mapping and scheduling problem : a List-based algorithm with backtracking, a Lookahead-based heuristic, a Bayesian-based heuristic and, a Q-Learning mapping algorithm. We evaluate and compare our solutions against an exhaustive approach in a real-life example, and illustrate the benefits and efficiency of the proposed framework.


Bien à vous,

Jean Cousty, Rostom Kachouri, Yukiko Kenmochi, Benjamin Perret et David Picard

pour la coordination des séminaires A3SI

<- Back to: Accueil