Abstract: Describing persons and their actions is a challenging problem due to variations in pose, scale and viewpoint in real-world images. Recently, semantic pyramids approach [1] for pose normalization has shown to provide excellent results for gender and action recognition. The performance of semantic pyramids approach relies on robust image description and is therefore limited due to the use of shallow local features. In the context of object recognition [2] and object detection [3], convolutional neural networks (CNNs) or deep features have shown to improve the performance over the conventional shallow features.
We propose deep semantic pyramids for human attributes and action recognition. The method works by constructing spatial pyramids based on CNNs of different part locations. These pyramids are then combined to obtain a single semantic representation. We validate our approach on the Berkeley and 27 Human Attributes datasets for attributes classification. For action recognition, we perform experiments on two challenging datasets: Willow and PASCAL VOC 2010. The proposed deep semantic pyramids provide a significant gain of 17.2%, 13.9%, 24.3% and 22.6% compared to the standard shallow semantic pyramids on Berkeley, 27 Human Attributes, Willow and PASCAL VOC 2010 datasets respectively. Our results also show that deep semantic pyramids outperform conventional CNNs based on the full bounding box of the person. Finally, we compare our approach with state-of-the-art methods and show a gain in performance compared to best methods in literature.

Abstract: Object detection with deep neural networks is often performed by passing a few
thousand candidate bounding boxes through a deep neural network for each image.
These bounding boxes are highly correlated since they originate from the same
image. In this paper we investigate how to exploit feature occurrence at the image scale to prune the neural network which is subsequently applied to all bounding boxes. We show that removing units which have near-zero activation in the image allows us to significantly reduce the number of parameters in the network. Results on the PASCAL 2007 Object Detection Challenge demonstrate that up to 40% of units in some fully-connected layers can be entirely eliminated with little change in the detection result.

Abstract: Part detection is an important aspect of object recognition. Most approaches apply object proposals to generate hundreds of possible part bounding box candidates which are then evaluated by part classifiers. Recently several methods have investigated directly regressing to a limited set of bounding boxes from deep neural network representation. However, for object parts such methods may be unfeasible due to their relatively small size with respect to the image. We propose a hierarchical method for object and part detection. In a single network we first detect the object and then regress to part location proposals based only on the feature representation inside the object. Experiments show that our hierarchical approach outperforms a network which directly regresses the part locations. We also show that our approach obtains part detection accuracy comparable or better than state-of-the-art on the CUB-200 bird and Fashionista clothing item datasets with only a fraction of the number of part proposals.

Abstract: This paper describes the monomodal and multimodal Neural Machine Translation systems developed by LIUM and CVC for WMT17 Shared Task on Multimodal Translation. We mainly explored two multimodal architectures where either global visual features or convolutional feature maps are integrated in order to benefit from visual context. Our final systems ranked first for both En-De and En-Fr language pairs according to the automatic evaluation metrics METEOR and BLEU.

Abstract: Ensembles are a popular way to improve results of discriminative CNNs. The
combination of several networks trained starting from different initializations
improves results significantly. In this paper we investigate the usage of ensembles of GANs. The specific nature of GANs opens up several new ways to construct ensembles. The first one is based on the fact that in the minimax game which is played to optimize the GAN objective the generator network keeps on changing even after the network can be considered optimal. As such ensembles of GANs can be constructed based on the same network initialization but just taking models which have different amount of iterations. These so-called self ensembles are much faster to train than traditional ensembles. The second method, called cascade GANs, redirects part of the training data which is badly modeled by the first GAN to another GAN. In experiments on the CIFAR10 dataset we show that ensembles of GANs obtain model probability distributions which better model the data distribution. In addition, we show that these improved results can be obtained at little additional computational cost.

Abstract: Hyperspectral signal reconstruction aims at recovering the original spectral input that produced a certain trichromatic (RGB) response from a capturing device or observer.
Given the heavily underconstrained, non-linear nature of the problem, traditional techniques leverage different statistical properties of the spectral signal in order to build informative priors from real world object reflectances for constructing such RGB to spectral signal mapping. However,
most of them treat each sample independently, and thus do not benefit from the contextual information that the spatial dimensions can provide. We pose hyperspectral natural image reconstruction as an image to image mapping learning problem, and apply a conditional generative adversarial framework to help capture spatial semantics. This is the first time Convolutional Neural Networks -and, particularly, Generative Adversarial Networks- are used to solve this task. Quantitative evaluation shows a Root Mean Squared Error (RMSE) drop of 44:7% and a Relative RMSE drop of 47:0% on the ICVL natural hyperspectral image dataset.

Abstract: In this paper we propose a multi-modal deep learning approach to detect floods in social media posts. Social media posts normally contain some metadata and/or visual information, therefore in order to detect the floods we use this information. The model is based on a Convolutional Neural Network which extracts the visual features and a bidirectional Long Short-Term Memory network to extract the semantic features from the textual metadata. We validate the
method on images extracted from Flickr which contain both visual information and metadata and compare the results when using both, visual information only or metadata only. This work has been done in the context of the MediaEval Multimedia Satellite Task.