The aesthetic quality of an image is defined as the measure or appreciation of the beauty of an image. Aesthetics is inherently a subjective property but there are certain factors that influence it such as, the semantic content of the image, the attributes describing the artistic aspect, the photographic setup used for the shot, etc. In this paper we propose a method for the automatic prediction of the aesthetics of an image that is based on the analysis of the semantic content, the artistic style and the composition of the image. The proposed network includes: a pre-trained network for semantic features extraction (the Backbone); a Multi Layer Perceptron (MLP) network that relies on the Backbone features for the prediction of image attributes (the AttributeNet); a self-adaptive Hypernetwork that exploits the attributes prior encoded into the embedding generated by the AttributeNet to predict the parameters of the target network dedicated to aesthetic estimation (the AestheticNet). Given an image, the proposed multi-network is able to predict: style and composition attributes, and aesthetic score distribution. Results on three benchmark datasets demonstrate the effectiveness of the proposed method, while the ablation study gives a better understanding of the proposed network.

With the rise in use of social media to promote branded products, the demand for effective influencer marketing has increased. Brands are looking for improved ways to identify valuable influencers among a vast catalogue; this is even more challenging with "micro-influencers", which are more affordable than mainstream ones but difficult to discover. In this paper, we propose a novel multi-task learning framework to improve the state of the art in micro-influencer ranking based on multimedia content. Moreover, since the visual congruence between a brand and influencer has been shown to be good measure of compatibility, we provide an effective visual method for interpreting our models’ decisions, which can also be used to inform brands’ media strategies. We compare with the current state-of-the-art on a recently constructed public dataset and we show significant improvement both in terms of accuracy and model complexity. The techniques for ranking and interpretation presented in this work can be generalised to arbitrary multimedia ranking tasks that have datasets with a similar structure.

In the last decade, an active area of research has been devoted to design novel activation functions that are able to help deep neural networks to converge, obtaining better performance. The training procedure of these architectures usually involves optimization of the weights of their layers only, while non-linearities are generally pre-specified and their (possible) parameters are usually considered as hyper-parameters to be tuned manually. In this paper, we introduce two approaches to automatically learn different combinations of base activation functions (such as the identity function, ReLU, and tanh) during the training phase. We present a thorough comparison of our novel approaches with well-known architectures (such as LeNet-5, AlexNet, and ResNet-56) on three standard datasets (Fashion-MNIST, CIFAR-10, and ILSVRC-2012), showing substantial improvements in the overall performance, such as an increase in the top-l accuracy for AlexNet on ILSVRC-2012 of 3.01 percentage points.

In this work we present a method to improve the pruning step of the current state-of-the-art methodology to compress neural networks. The novelty of the proposed pruning technique is in its differentiability, which allows pruning to be performed during the backpropagation phase of the network training. This enables an end-to-end learning and strongly reduces the training time. The technique is based on a family of differentiable pruning functions and a new regularizer specifically designed to enforce pruning. The experimental results show that the joint optimization of both the thresholds and the network weights permits to reach a higher compression rate, reducing the number of weights of the pruned network by a further 14% to 33 % compared to the current state-of-the-art. Furthermore, we believe that this is the first study where the generalization capabilities in transfer learning tasks of the features extracted by a pruned network are analyzed. To achieve this goal, we show that the representations learned using the proposed pruning methodology maintain the same effectiveness and generality of those learned by the corresponding non-compressed network on a set of different recognition tasks.

We present a theoretically grounded approach to train deep neural networks, including recurrent networks, subject to class-dependent label noise. We propose two procedures for loss correction that are agnostic to both application domain and network architecture. They simply amount to at most a matrix inversion and multiplication, provided that we know the probability of each class being corrupted into another. We further show how one can estimate these probabilities, adapting a recent technique for noise estimation to the multi-class setting, and thus providing an end-to-end framework. Extensive experiments on MNIST, IMDB, CIFAR-10, CIFAR-100 and a large scale dataset of clothing images employing a diversity of architectures - stacking dense, convolutional, pooling, dropout, batch normalization, word embedding, LSTM and residual layers - demonstrate the noise robustness of our proposals. Incidentally, we also prove that, when ReLU is the only non-linearity, the loss curvature is immune to class-dependent label noise.

Motion detection in video streams is a challenging task for several computer vision applications. Indeed, segmentation of moving and static elements in the scene allows to increase the efficiency of several challenging tasks, such as human-computer interface, robot visions, and intelligent surveillance systems. In this paper, we approach motion detection through a multi-layered artificial neural network, which is able to build for each background pixel a multi-modal color distribution evolving over time through self-organization. According to the winner-take-all rule, each layer of the network models an independent state of the background scene, in response to external disturbing conditions, such as illumination variations, moving backgrounds, and jittering. As a result, our background subtraction method exhibits high generalization capabilities that in combination with a post-processing filtering schema allow to produce accurate motion segmentation. Moreover, we propose an approach to detect anomalous events (such as camera motion) that require background model re-initialization. We describe our method in full details and we compare it against the most recent background subtraction approaches. Experimental results for video sequences from the 2012 and 2014 CVPR Change Detection data sets demonstrate how our methodology outperforms many state-of-the-art methods in terms of detection rate.

State-of-the-art background subtraction approaches exhibit impressive results in several applications, however, they usually employ static parameter values that reduce their flexibility when used in real world environments. In this paper we relax this problem by proposing a novel approach based on temporal analysis for automatic adjustment of the parameter that controls the motion segmentation quality. Moreover, we extend and strengthen the Self Organizing map Background Subtraction (SOBS) into a multi-layered framework defining a multi-modal background color representation. According to the winner-take-all rule, each layer builds a background model representing an independent view of the scene as response to external conditions such as illumination variations, camouflages, and moving backgrounds. Experimental results on the 2012 CVPR Change Detection dataset, demonstrate that our approach outperforms SOBS achieving promising results compared with those obtained by state-of-the-art background subtraction methodologies.

This paper reports a novel approach to deal with the problem of Object and Scene recognition extending the traditional Bag of Words approach in two ways. Firstly, a dataset independent method of summarizing local features, based on multivariate Gaussian descriptors, is employed. Secondly, a recently proposed classification technique, particularly suited for high dimensional feature spaces without any dimensionality reduction step, allows to effectively exploit these features. Experiments are performed on two publicly available datasets and demonstrate the effectiveness of our approach when compared to state-of-the-art methods.

Graph-based semi-supervised learning methods play a key role in machine learning applications, particularly when no parametric information or other prior knowledge is available. Given a graph whose nodes represent the points and the weighted edges the relations between them, the goal is to predict the values of all unlabeled nodes exploiting the information provided by both label and unlabeled nodes. In this paper, we propose a novel graph-based approach for semi-supervised binary classification. The algorithm extends the Fisher Subspace estimation approaches by adopting a kernel graph covariance measure. This similarity measure defines a relation between nodes generalizing both the shortest path and the commute time distance. This quantity is called the sum-over-paths covariance. Experiments on synthetic and real-world datasets highlight that the proposed algorithm achieves better results with respect to those obtained by state-of-the-art competitors.

This paper proposes a novel two-class classifier, called IPCAC, based on the isotropic principal component analysis technique; it allows to deal with training data drawn from mixture of Gaussian distributions, by projecting the data on the Fisher subspace that separates the two classes. The obtained results demonstrate that IPCAC is a promising technique; furthermore, to cope with training datasets being dynamically supplied, and to work with non-linearly separable classes, two improvements of this classifier are defined: a model merging algorithm, and a kernel version of IPCAC. The effectiveness of the proposed methods is shown by their application to the spam classification problem, and by the comparison of the achieved results with those obtained by support vector machines SVM, and K-nearest neighbors KNN.