Mo and Morgado introduced in 2022 a novel self-supervised learning approach for Visual Sound Source Localization, denoted as SLAVC [Mo, S. and Mordado, P., A Closer Look at Weakly-Supervised Audio-Visual Source Localization, Advances in Neural Information Processing Systems, 2022]. The proposed method is based on multiple-instance contrastive learning. In addition to improving the results of previous methods, it also solves two critical problems that former methods faced: 1) excessive overfitting despite training on extensive datasets, 2) tendency to hallucinate sound sources even without visual evidence to support it in the video. In this paper, we briefly present the method, offer an online executable version allowing the users to test it on their own image-audio pairs and propose some improvements that could benefit the model as future work.

International audience

Federated Learning (FL) is a learning paradigm where multiple nodes collaboratively train a model by only exchanging updates or parameters. This enables to keep data locally, therefore enhancing privacy - statement requiring nuance, e.g. memorization of training data in language models. Depending on the application, the number of samples that each node contains can be very different, which can impact the training and the final performance. This work studies the impact of the per-node sample size distribution on the mean squared error (MSE) of the one-shot federated estimator. We focus on one-shot aggregation of statistical estimations made across disjoint, independent and identically distributed (i.i.d.) data sources, in the context of empirical risk minimization. In distributed learning, it is well-known that for a total number of m nodes, each node should contain at least m samples to equal the performance of centralized training. In a federated scenario, this result remains true, but now applies to the mean of the per-node sample size distribution. The demo enables to visualize this effect as well as to compare the behavior of the FESC (Federated Estimation with Statistical Correction) algorithm - a weighting scheme which depends on the local sample size - with respect to the classical federated estimator and the centralized one, for a large collection of distributions, number of nodes, and features space dimension. **This is an MLBriefs article, the source code has not been reviewed!**<br> **The original source code is [[available here|https://github.com/gonzaq94/one-shot-fed-learning]] (last checked 2023/02/12).**

Binary shapes, or silhouettes, are building elements of logos, graphic symbols and fonts， which require various forms of geometric editing without compromising the resolution. In this paper, we present an effective silhouette vectorization algorithm that extracts the outline of a 2D shape from a raster binary image and converts it to a combination of cubic Bézier polygons and perfect circles. Compared to state-of-the-art image vectorization software, this algorithm has demonstrated a superior reduction in the number of control points while maintaining high accuracy.

Time series of satellite images are now massively available thanks to the existence of several constellations of recurrent satellites. We propose a method for detecting and measuring the duration of changes on such series. This approach is intended to be generic and independent of the type of satellite used, whether band limited or multispectral. It is based on a global analysis of the sequence. The statistical detection method is applied to a residual sequence computed from backward and forward novelty filters applied to all images in the series. Significant changes are computed with a guarantee on their number of false alarms (NFA). To establish the efficiency of the method, we have created an open database of 28 sequences of 20 images acquired by the Sentinel-2 satellite, in different regions of the world. We obtain satisfactory results, which are consistent with the visual observations.

In this work, we propose an open-source Python implementation of the SAX-ARM algorithm introduced by Park and Jung (2019). This algorithm mines association rules efficiently among the deviant events of multivariate time series. To do so, the algorithm combines two existing methods, namely the Symbolic Aggregate approXimation (SAX) from Lin et al. (2003) - a symbolic representation of time series - and the Apriori algorithm from Agrawal et al. (1996) - a data mining method which outputs all frequent itemsets and association rules from a transactional dataset. A detailed description of the underlying principles is given along with their numerical implementation. The choice of relevant parameters is thoroughly discussed and evaluated using a public dataset on the topic of temperature and energy consumption.

The algorithm presented in this paper is an application of a general framework for morphological processing of signals on weighted graphs. Here we apply it to images by defining what we call a co-circularity graph. In this graph, the vertices are the pixels and the weighted edges depend on a consistency criterion (co-circularity) between local orientations estimated from the structure tensors. This graph induces anisotropic adaptive morphological operators which are related both to anisotropic diffusion in images and path optimality in graphs. We present several applications such as the enhancement of fiber-like structures, completion of interrupted edges and the regularization of grayscale images. We also discuss the parameters setting depending on the application.

In computer vision, and particularly in 3D reconstruction from images, it is customary to be faced with regression problems contaminated by outlying data. The standard and efficient method to deal with them is the Random Sample Consensus (RANSAC) algorithm. The procedure is simple and versatile, drawing random minimal samples from the data to estimate parameterized candidate models and ranking them based on the amount of compatible data. Such evaluation involves some threshold that separates inliers from outliers. In presence of unknown level of noise, as is usual in practice, it is desirable to remove the dependency on this fixed threshold and to estimate it as an additional unknown. Among the numerous variants of RANSAC, few, that we call 'automatic', propose this approach, which involves changing the maximization criterion of consensus, as it is naturally increasing with the varying threshold. An algorithm of Zach and Cohen (ICCV 2015) uses the likelihood statistics. We present the details and the implementation of their method along with quantitative and qualitative tests on standard stereovision tasks: estimation of homography, fundamental and essential matrix.

VBM3D is an extension to video of the well-known image denoising algorithm BM3D, which takes advantage of the sparse representation of stacks of similar patches in a transform domain. The extension is rather straightforward: the similar 2D patches are taken from a spatio-temporal neighborhood which includes neighboring frames. In spite of its simplicity, the algorithm offers a good trade-off between denoising performance and computational complexity. In this work we revisit this method, providing an open-source C++ implementation reproducing the results. A detailed description is given and the choice of parameters is thoroughly discussed. Furthermore, we study and compare several extensions of the original algorithm: (1) a multiscale implementation, (2) the use of 3D patches, (3) the use of optical flow to guide the patch search. These extensions yield results which are competitive with the most recent state of the art.