The IEEE Conference on Computer Vision and Pattern Recognition 2017, CVPR ‘2017 , is the premier annual computer vision event comprising the main conference and several co-located workshops, the best meeting point for students, academics and industry researchers in this area, where Deep Learning is gaining momentum.  This year the conference will be held between July 21-26, , in Honolulu (Hawaii, USA).

Our research group at BSC-CNS  and UPC Barcelona Tech will be present at CVPR’2017 with contribution in two workshops. I’m happy to have this awesome opportunity, thanks to our collaboration with the research group at UPC Barcelona Tech  led by  Xavier Giró i Nieto.  We hope that with our effort and collaboration, we can help to turn Barcelona into an Deep Learning hub.  Throughout the next academic year, we will promote different initiatives in this direction. For now, we are going to start in September with a new Deep Learning for Artificial Intelligence master course at ETSETB (UPC).

In this direction, as far as I know, there are  three other contributions to CVPR’17 from UPC members. Two from my colleagues at UPC Francesc Moreno-Noguer and Antonio Agudo  with the papers 3D Human Pose Estimation from a Single Image via Distance Matrix Regression and DUST: Dual Union of Spatio-Temporal Subspaces for Monocular Multiple Object 3D Reconstruction.  Another paper will by presented by Amaia Salvador titled Learning Cross-Modal Embeddings for Cooking Recipes and Food Images.

Also other colleagues from Barcelona will also present their work in this conference. As far as I know, Àgata Lapedriza Garcia with the paper  Emotion Recognition in Context. Carles Ventura, David Masip and also Àgata with the paper Interpreting CNN Models for Apparent Personality Trait Regression. And Sergio Escalera who is the organiser of the ChaLearn Looking at People series Workshop and challengePlease, let me know if I leave anyone on this list (at torres@ac.upc.edu). I would appreciate it in order to update this post. I will be happy doing it.

And to finish, here is a brief description of our humble work we have done in collaboration with other institutions to whom I appreciate the opportunity they have granted me.

Disentangling Motion, Foreground and Background Features in Videos
In this work we present a novel architecture for unsupervised video features learning by disentangling motion, foreground and background features in videos. Supervised preliminary experiments were conducted to prove the feasibility of our proposal. Our work provides better-generalized video features than other methods in the discriminative task like action recognition.

Our model and source code are publicly available at this https URL  The paper that contains the detailed contribution can be downloaded from arxiv repository.

This first work was developed mainly during an exchange period of Xunyu Lin at the Universitat Politecnica de Catalunya with the collaboration of  Victor Campos, Xavier Giró, Jordi Torres and Cristian Canton from the following institutions:  Universitat Politècnica de Catalunya, Beihang University, Barcelona Supercomputing Center and Facebook.

 

SalGAN: Visual Saliency Prediction with Generative Adversarial Networks

In this second work we introduce SalGAN, a deep convolutional neural network for visual saliency prediction trained with adversarial examples. The first stage of the network consists of a generator model whose weights are learned by back-propagation computed from a binary cross entropy (BCE) loss over downsampled versions of the saliency maps. The resulting prediction is processed by a discriminator network trained to solve a binary classification task between the saliency maps generated by the generative stage and the ground truth ones. Our experiments show how adversarial training allows reaching state-of-the-art performance across different metrics when combined with a widely-used loss function like BCE.

Our model and source code are publicly available at this https URL .  The shorter extended abstract presented as spotlight in the CVPR 2017 Scene Understanding Workshop (SUNw) is available here.
This first work was developed in joint collaboration between:

 

Junting Pan Cristian Canton Ferrer Kevin McGuinness Noel O'Connor Jordi Torres Elisa Sayrol Xavier Giro-i-Nieto
Junting Pan Cristian Canton Ferrer Kevin McGuinness Noel O’Connor Jordi Torres Elisa Sayrol Xavier Giro-i-Nieto
logo-insight logo-dcu logo-microsoft
Insight Centre for Data Analytics Dublin City University (DCU) Microsoft
logo-facebook logo-bsc logo-upc
Facebook Barcelona Supercomputing Center Universitat Politecnica de Catalunya (UPC)
2017-08-08T13:21:22+00:00 July 14th, 2017|