Conference Program

Content based multichannel bioacoustic indexing & autoencoder for 3D whale survey
in collaboration with Phd students Marion Poupard and Ferrari Maxence, CNRS LIS Toulon and with Seaproven SA

In this talk, we first recall the main objectives of massive bioacoustic indexing.
We will present usual and more advanced stereophonic detection of Whales from long term submarine observatory.
We demonstrate algorithms working on bottom mounted hydrophone antenna or fixed under mobile autonomous vehicles to computed detailed 3D tracks of the dive of the whale during hours.
We then propose a new paradigm towards deep learning of multichannel bioacoustic regresentation from raw audio. We thus take advantages of stereophony (or multiphony)
to elaborate new objective loss function for transient detection, denoising and accurate estimation Time Delay of Arrival (TDOA).
We train autoencoder on stereo transients from whale recordings, and show how to improve TDOA estimation taking advantages of estimator from learned limited development of a transfer function.
We then discuss perspectives towards embedded deep learning method in autonomous vehicle to index biodiversity of the abyss.

Session chair: Andrea Manno-Kovacs, SZTAKI

Saliency Weighted Convolutional Features for Instance Search
Eva Mohedano, Kevin McGuinness, Xavier Giro-I-Nieto and Noel O'Connor

Introduction of Explicit Visual Saliency in Training of Deep CNNs: Application to Architectural Styles Classification
Abraham Montoya Obeso, Jenny Benois-Pineau, Mireya S. Garcia Vazquez and Alejandro Alvaro Ramirez Acosta

Interestingness for Immersive Content: Prediction and Use
Gwenaelle Marquant, Claire-Hélène Demarty, Christel Chamaret, Joel Sirot and Louis Chevallier

Refining and Evaluating a Video Sharpness Metric
Stefanie Wechtitsch, Werner Bailer and Lucas Paletta

UNNA: A Unified Neural Network for Aesthetic Assessment
Larbi Abdenebaoui, Benjamin Meyer, Albert Bruns and Susanne Boll

Session chair: Alan HANJALIC, TU Delft

Latent Feature Combination for Multi-Context Music Recommendation
Martin Pichl and Eva Zangerle

Detection and Localization of Partial Audio Matches
Milica Maksimović, Patrick Aichroth and Luca Cuccovillo

Mastering Occlusions by Using Intelligent Facial Frameworks Based on the Propagation of Movement
Delphine Poux, Benjamin Allaert, José Mennesson, Nacim Ihaddadene, Ioan Marius Bilasco and Chaabane Djeraba

Deep Multimodal Features for Movie Genre and Interestingness Prediction
Olfa Ben Ahmed and Benoit Huet

Ridiculously Fast Shot Boundary Detection with Fully Convolutional Neural Networks
Michael Gygli

CAD-SE: Discovering CAD Models on the Web
Christian Brosch, Andreas Seifert and Thomas Köllmer

DeCP-Live: A Web-Interface for DeCP, a Distributed High-Throughput CBIR System
Gylfi Þór Guðmundsson, Christian Andreas Jacobsen, Hilmar Tryggvason and Björn Þór Jónsson

Musical Gesture Recognition Using Machine Learning and Audio Descriptors
Paul Best, Jean Bresson and Diemo Schwarz

Dynamic Index Finger Gesture Video Dataset for Mobile Interaction
Cagan Arslan, Ioan Marius Bilasco and Jean Martinet

Boat trip on l’Espérance in the Bay of La Rochelle and around the Ré Island.
The boat l’Espérance is located in the old port of La Rochelle, at 10 minutes walk from the conference venue (see map below).

Enjeux socio-politiques de l’IA (Socio-political issues of AI)

Profitant des réussites spectaculaires des techniques de deep learning, les promesses de l’Intelligence artificielle (IA) sont revenues occuper le débat public. Dans cette conférence, on propose de retracer quelques aspects de l’histoire de l’Intelligence artificielle afin de comprendre les défis présents. Le « troisième printemps de l’IA » se caractérise, en effet, par le retour du paradigme connexionniste qui avait été marginalisé lors du « second printemps » de l’Intelligence artificielle dans les années quatre-vingt. La nouvelle vague de promesses de l’IA a pris forme grâce aux opportunités nouvelles offertes par les données massives et aux nouvelles capacités de calcul des ordinateurs. Mais, elle propose surtout une autre manière de rendre les machines « intelligentes ». Il ne s’agit plus de demander aux calculateurs de raisonner, mais bien d’apprendre des données et de former des modèles de prédiction à partir des données. Ce changement de paradigme de l’IA a de nombreuses conséquences : il propose une représentation différentes de la société qui s’appuie sur le comportement des individus plutôt que sur leurs catégories d’appartenance, il ouvre des possibilités d’automatisation du travail, il présente des opportunités nouvelles mais aussi des risques de biais dans les calculs. Cette conférence propose d’examiner les défis éthiques et politiques que pose les développements de l’IA.

Gala Dinner aboard l’Espérance boat.
L’Espérance is located in the old port of La Rochelle, at 10 minutes walk from the conference venue (see map below).

BigEarth-Accurate and Scalable Processing of Big Data in Earth Observation
Begüm Demir

Urban Land Use Classification Based on Aerial and Ground Images
Rui Cao and Guoping Qiu

Overcoming Missing and Incomplete Modalities with Generative Adversarial Networks for Building Footprint Segmentation
Benjamin Bischke, Patrick Helber, Florian Koenig, Damian Borth and Andreas Dengel

A Study on the Sensitivity-Based Discriminative Hyperspectral Image Content Representation
Olfa Ben Ahmed, Thierry Urruty, Noel Richard and Christine Fernandez

Active Learning to Assist Annotation of Aerial Images in Environmental Surveys
Mathieu Laroze, Romain Dambreville, Chloé Friguet, Ewa Kijak and Sébastien Lefèvre

Session chairs: Sébastien Lefèvre, Université de Bretagne Sud and Josiane Mothe, IRIT Toulouse

Coupled Ensembles of Neural Networks
Anuvabh Dutt, Denis Pellerin and Georges Quénot

Temporal Compression and Fast Matching of Hand-crafted and Deep Features of Video Segments
Werner Bailer and Stefanie Wechtitsch

Sport Action Recognition with Siamese Spatio-Temporal CNNs: Application to Table Tennis
Pierre-Etienne Martin, Jenny Benois-Pineau, Renaud Peteri and Julien Morlier

Deep 3D Flow Features for Human Action Recognition
Athanasios Psaltis, Georgios Papadopoulos and Petros Daras

Action Tube Extraction based 3D-CNN for RGB-D Action Recognition
Zineng Xu, Veronica Vilaplana and Josep Ramon Morros

Session chair: Georges Quénot, LIG-CNRS