Radu Horaud

Radu Horaud, Fadi Dornaika

Whenever a sensor is mounted on a robot hand it is important to know the relationship between the sensor and the hand. The problem of determining this relationship is referred to as hand-eye calibration, which is important in at least two types of tasks: (i) map sensor centered measurements into the robot workspace and (ii) allow the robot to precisely move the sensor. In the past some solutions were proposed in the particular case of a camera. With almost no exception, all existing solutions attempt to solve the homogeneous matrix equation AX=XB. First we show that there are two possible formulations of the hand-eye calibration problem. One formulation is the classical one that we just mentioned. A second formulation takes the form of the following homogeneous matrix equation: MY=M'YB. The advantage of the latter is that the extrinsic and intrinsic camera parameters need not be made explicit. Indeed, this formulation directly uses the 3 by 4 perspective matrices (M and M') associated with two positions of the camera. Moreover, this formulation together with the classical one cover a wider range of camera-based sensors to be calibrated with respect to the robot hand. Second, we develop a common mathematical framework to solve for the hand-eye calibration problem using either of the two formulations. We present two methods, (i) a rotation then translation and (ii) a non-linear solver for rotation and translation. Third, we perform a stability analysis both for our two methods and for the classical linear method of Tsai and Lenz (1989). In the light of this comparison, the non-linear optimization method, that solves for rotation and translation simultaneously, seems to be the most robust one with respect to noise and to measurement errors.

Nicolas Andreff, Radu Horaud, Bernard Espiau

In this paper we propose a new flexible method for hand-eye calibration. The vast majority of existing hand-eye calibration techniques requires a calibration rig which is used in conjunction with camera pose estimation methods. Instead, we combine structure-from-motion with known robot motions and we show that the solution can be obtained in linear form. The latter solves for both the hand-eye parameters and for the unknown scale factor inherent with structure-from-motion methods. The algebraic analysis that is made possible with such a linear formulation allows to investigate not only the well known case of general screw motions but also such singular motions as pure translations, pure rotations, and planar motions. In essence, the robot-mounted camera looks to an unknown rigid layout, tracks points over an image sequence and estimates the camera-to-robot relationship. Such a self calibration process is relevant for unmanned vehicles, robots working in remote places, and so forth. We conduct a large number of experiments which validate the quality of the method by comparing it with existing ones.

Radu Horaud, Fadi Dornaika, Bernard Espiau

In this paper we present a visual servoing approach to the problem of object grasping and more generally, to the problem of aligning an end-effector with an object. First we extend the method proposed by Espiau et al. [1] to the case of a camera which is not mounted onto the robot being controlled and we stress the importance of the real-time estimation of the image Jacobian. Second, we show how to represent a grasp or more generally, an alignment between two solids in 3-D projective space using an uncalibrated stereo rig. Such a 3-D projective representation is view-invariant in the sense that it can be easily mapped into an image set-point without any knowledge about the camera parameters. Third, we perform an analysis of the performances of the visual servoing algorithm and of the grasping precision that can be expected from this type of approach.

Radu Horaud, Humberto Sossa

In computer vision, the indexing problem is the problem of recognizing a few objects in a large database of objects while avoiding the help of the classical image-feature-to-object-feature matching paradigm. In this paper we address the problem of recognizing 3-D polyhedral objects from 2-D images by indexing. Both the objects to be recognized and the images are represented by weighted graphs. The indexing problem is therefore the problem of determining whether a graph extracted from the image is present or absent in a database of model graphs. We introduce a novel method for performing this graph indexing process which is based both on polynomial characterization of binary and weighted graphs and on hashing. We describe in detail this polynomial characterization and then we show how it can be used in the context of polyhedral object recognition. Next we describe a practical recognition-by-indexing system that includes the organization of the database, the representation of polyhedral objects in terms of 2-D characteristic views, the representation of this views in terms of weighted graphs, and the associated image processing. Finally, some experimental results allow the evaluation of the system performance.

Zhiqi Kang, Mostafa Sadeghi, Radu Horaud et al.

Face frontalization consists of synthesizing a frontally-viewed face from an arbitrarily-viewed one. The main contribution of this paper is a frontalization methodology that preserves non-rigid facial deformations in order to boost the performance of visually assisted speech communication. The method alternates between the estimation of (i)~the rigid transformation (scale, rotation, and translation) and (ii)~the non-rigid deformation between an arbitrarily-viewed face and a face model. The method has two important merits: it can deal with non-Gaussian errors in the data and it incorporates a dynamical face deformation model. For that purpose, we use the generalized Student t-distribution in combination with a linear dynamic system in order to account for both rigid head motions and time-varying facial deformations caused by speech production. We propose to use the zero-mean normalized cross-correlation (ZNCC) score to evaluate the ability of the method to preserve facial expressions. The method is thoroughly evaluated and compared with several state of the art methods, either based on traditional geometric models or on deep learning. Moreover, we show that the method, when incorporated into deep learning pipelines, namely lip reading and speech enhancement, improves word recognition and speech intelligibilty scores by a considerable margin. Supplemental material is accessible at https://team.inria.fr/robotlearn/research/facefrontalization/

Yihong Xu, Aljosa Osep, Yutong Ban et al.

The recent trend in vision-based multi-object tracking (MOT) is heading towards leveraging the representational power of deep learning to jointly learn to detect and track objects. However, existing methods train only certain sub-modules using loss functions that often do not correlate with established tracking evaluation measures such as Multi-Object Tracking Accuracy (MOTA) and Precision (MOTP). As these measures are not differentiable, the choice of appropriate loss functions for end-to-end training of multi-object tracking methods is still an open research problem. In this paper, we bridge this gap by proposing a differentiable proxy of MOTA and MOTP, which we combine in a loss function suitable for end-to-end training of deep multi-object trackers. As a key ingredient, we propose a Deep Hungarian Net (DHN) module that approximates the Hungarian matching algorithm. DHN allows estimating the correspondence between object tracks and ground truth objects to compute differentiable proxies of MOTA and MOTP, which are in turn used to optimize deep trackers directly. We experimentally demonstrate that the proposed differentiable framework improves the performance of existing multi-object trackers, and we establish a new state of the art on the MOTChallenge benchmark. Our code is publicly available from https://github.com/yihongXU/deepMOT.

Xavier Alameda-Pineda, Angus Addlesee, Daniel Hernández García et al.

Despite the many recent achievements in developing and deploying social robotics, there are still many underexplored environments and applications for which systematic evaluation of such systems by end-users is necessary. While several robotic platforms have been used in gerontological healthcare, the question of whether or not a social interactive robot with multi-modal conversational capabilities will be useful and accepted in real-life facilities is yet to be answered. This paper is an attempt to partially answer this question, via two waves of experiments with patients and companions in a day-care gerontological facility in Paris with a full-sized humanoid robot endowed with social and conversational interaction capabilities. The software architecture, developed during the H2020 SPRING project, together with the experimental protocol, allowed us to evaluate the acceptability (AES) and usability (SUS) with more than 60 end-users. Overall, the users are receptive to this technology, especially when the robot perception and action skills are robust to environmental clutter and flexible to handle a plethora of different interactions.

Zhiqi Kang, Mostafa Sadeghi, Radu Horaud et al.

This paper investigates the impact of head movements on audio-visual speech enhancement (AVSE). Although being a common conversational feature, head movements have been ignored by past and recent studies: they challenge today's learning-based methods as they often degrade the performance of models that are trained on clean, frontal, and steady face images. To alleviate this problem, we propose to use robust face frontalization (RFF) in combination with an AVSE method based on a variational auto-encoder (VAE) model. We briefly describe the basic ingredients of the proposed pipeline and we perform experiments with a recently released audio-visual dataset. In the light of these experiments, and based on three standard metrics, namely STOI, PESQ and SI-SDR, we conclude that RFF improves the performance of AVSE by a considerable margin.

Vineet Gandhi, Jan Cech, Radu Horaud

The combination of range sensors with color cameras can be very useful for robot navigation, semantic perception, manipulation, and telepresence. Several methods of combining range- and color-data have been investigated and successfully used in various robotic applications. Most of these systems suffer from the problems of noise in the range-data and resolution mismatch between the range sensor and the color cameras, since the resolution of current range sensors is much less than the resolution of color cameras. High-resolution depth maps can be obtained using stereo matching, but this often fails to construct accurate depth maps of weakly/repetitively textured scenes, or if the scene exhibits complex self-occlusions. Range sensors provide coarse depth information regardless of presence/absence of texture. The use of a calibrated system, composed of a time-of-flight (TOF) camera and of a stereoscopic camera pair, allows data fusion thus overcoming the weaknesses of both individual sensors. We propose a novel TOF-stereo fusion method based on an efficient seed-growing algorithm which uses the TOF data projected onto the stereo image pair as an initial set of correspondences. These initial "seeds" are then propagated based on a Bayesian model which combines an image similarity score with rough depth priors computed from the low-resolution range data. The overall result is a dense and accurate depth map at the resolution of the color cameras at hand. We show that the proposed algorithm outperforms 2D image-based stereo algorithms and that the results are of higher resolution than off-the-shelf color-range sensors, e.g., Kinect. Moreover, the algorithm potentially exhibits real-time performance on a single CPU.

Avinash Sharma, Radu Horaud, Diana Mateus

We address the problem of 3D shape registration and we propose a novel technique based on spectral graph theory and probabilistic matching. The task of 3D shape analysis involves tracking, recognition, registration, etc. Analyzing 3D data in a single framework is still a challenging task considering the large variability of the data gathered with different acquisition devices. 3D shape registration is one such challenging shape analysis task. The main contribution of this chapter is to extend the spectral graph matching methods to very large graphs by combining spectral graph matching with Laplacian embedding. Since the embedded representation of a graph is obtained by dimensionality reduction we claim that the existing spectral-based methods are not easily applicable. We discuss solutions for the exact and inexact graph isomorphism problems and recall the main spectral properties of the combinatorial graph Laplacian; We provide a novel analysis of the commute-time embedding that allows us to interpret the latter in terms of the PCA of a graph, and to select the appropriate dimension of the associated embedded metric space; We derive a unit hyper-sphere normalization for the commute-time embedding that allows us to register two shapes with different samplings; We propose a novel method to find the eigenvalue-eigenvector ordering and the eigenvector signs using the eigensignature (histogram) which is invariant to the isometric shape deformations and fits well in the spectral graph matching framework, and we present a probabilistic shape matching formulation using an expectation maximization point registration algorithm which alternates between aligning the eigenbases and finding a vertex-to-vertex assignment.

Andrei Zaharescu, Radu Horaud

In this paper we address the problem of building a class of robust factorization algorithms that solve for the shape and motion parameters with both affine (weak perspective) and perspective camera models. We introduce a Gaussian/uniform mixture model and its associated EM algorithm. This allows us to address robust parameter estimation within a data clustering approach. We propose a robust technique that works with any affine factorization method and makes it robust to outliers. In addition, we show how such a framework can be further embedded into an iterative perspective factorization scheme. We carry out a large number of experiments to validate our algorithms and to compare them with existing ones. We also compare our approach with factorization methods that use M-estimators.

Diana Mateus, Radu Horaud, David Knossow et al.

Matching articulated shapes represented by voxel-sets reduces to maximal sub-graph isomorphism when each set is described by a weighted graph. Spectral graph theory can be used to map these graphs onto lower dimensional spaces and match shapes by aligning their embeddings in virtue of their invariance to change of pose. Classical graph isomorphism schemes relying on the ordering of the eigenvalues to align the eigenspaces fail when handling large data-sets or noisy data. We derive a new formulation that finds the best alignment between two congruent $K$-dimensional sets of points by selecting the best subset of eigenfunctions of the Laplacian matrix. The selection is done by matching eigenfunction signatures built with histograms, and the retained set provides a smart initialization for the alignment problem with a considerable impact on the overall performance. Dense shape matching casted into graph matching reduces then, to point registration of embeddings under orthogonal transformations; the registration is solved using the framework of unsupervised clustering and the EM algorithm. Maximal subset matching of non identical shapes is handled by defining an appropriate outlier class. Experimental results on challenging examples show how the algorithm naturally treats changes of topology, shape variations and different sampling densities.

Radu Horaud, Miles Hansard, Georgios Evangelidis et al.

Time-of-flight (TOF) cameras are sensors that can measure the depths of scene-points, by illuminating the scene with a controlled laser or LED source, and then analyzing the reflected light. In this paper, we will first describe the underlying measurement principles of time-of-flight cameras, including: (i) pulsed-light cameras, which measure directly the time taken for a light pulse to travel from the device to the object and back again, and (ii) continuous-wave modulated-light cameras, which measure the phase difference between the emitted and received signals, and hence obtain the travel time indirectly. We review the main existing designs, including prototypes as well as commercially available devices. We also review the relevant camera calibration principles, and how they are applied to TOF devices. Finally, we discuss the benefits and challenges of combined TOF and color camera systems.

Georgios D. Evangelidis, Miles Hansard, Radu Horaud

This paper addresses the problem of range-stereo fusion, for the construction of high-resolution depth maps. In particular, we combine low-resolution depth data with high-resolution stereo data, in a maximum a posteriori (MAP) formulation. Unlike existing schemes that build on MRF optimizers, we infer the disparity map from a series of local energy minimization problems that are solved hierarchically, by growing sparse initial disparities obtained from the depth data. The accuracy of the method is not compromised, owing to three properties of the data-term in the energy function. Firstly, it incorporates a new correlation function that is capable of providing refined correlations and disparities, via subpixel correction. Secondly, the correlation scores rely on an adaptive cost aggregation step, based on the depth data. Thirdly, the stereo and depth likelihoods are adaptively fused, based on the scene texture and camera geometry. These properties lead to a more selective growing process which, unlike previous seed-growing methods, avoids the tendency to propagate incorrect disparities. The proposed method gives rise to an intrinsically efficient algorithm, which runs at 3FPS on 2.0MP images on a standard desktop computer. The strong performance of the new method is established both by quantitative comparisons with state-of-the-art methods, and by qualitative comparisons using real depth-stereo data-sets.

Miles Hansard, Radu Horaud

The geometry of binocular projection is analyzed, with reference to the primate visual system. In particular, the effects of coordinated eye movements on the retinal images are investigated. An appropriate oculomotor parameterization is defined, and is shown to complement the classical version and vergence angles. The midline horopter is identified, and subsequently used to construct the epipolar geometry of the system. It is shown that the Essential matrix can be obtained by combining the epipoles with the projection of the midline horopter. A local model of the scene is adopted, in which depth is measured relative to a plane containing the fixation point. The binocular disparity field is given a symmetric parameterization, in which the unknown scene-depths determine the location of corresponding image-features. The resulting Cyclopean depth-map can be combined with the estimated oculomotor parameters, to produce a local representation of the scene. The recovery of visual direction and depth from retinal images is discussed, with reference to the relevant psychophysical and neurophysiological literature.

Yves Dufournaud, Cordelia Schmid, Radu Horaud

In this paper we address the problem of matching two images with two different resolutions: a high-resolution image and a low-resolution one. The difference in resolution between the two images is not known and without loss of generality one of the images is assumed to be the high-resolution one. On the premise that changes in resolution act as a smoothing equivalent to changes in scale, a scale-space representation of the high-resolution image is produced. Hence the one-to-one classical image matching paradigm becomes one-to-many because the low-resolution image is compared with all the scale-space representations of the high-resolution one. Key to the success of such a process is the proper representation of the features to be matched in scale-space. We show how to represent and extract interest points at variable scales and we devise a method allowing the comparison of two images at two different resolutions. The method comprises the use of photometric- and rotation-invariant descriptors, a geometric model mapping the high-resolution image onto a low-resolution image region, and an image matching strategy based on local constraints and on the robust estimation of this geometric model. Extensive experiments show that our matching method can be used for scale changes up to a factor of 6.

Andrei Zaharescu, Edmond Boyer, Radu Horaud

Triangulated meshes have become ubiquitous discrete-surface representations. In this paper we address the problem of how to maintain the manifold properties of a surface while it undergoes strong deformations that may cause topological changes. We introduce a new self-intersection removal algorithm, TransforMesh, and we propose a mesh evolution framework based on this algorithm. Numerous shape modelling applications use surface evolution in order to improve shape properties, such as appearance or accuracy. Both explicit and implicit representations can be considered for that purpose. However, explicit mesh representations, while allowing for accurate surface modelling, suffer from the inherent difficulty of reliably dealing with self-intersections and topological changes such as merges and splits. As a consequence, a majority of methods rely on implicit representations of surfaces, e.g. level-sets, that naturally overcome these issues. Nevertheless, these methods are based on volumetric discretizations, which introduce an unwanted precision-complexity trade-off. The method that we propose handles topological changes in a robust manner and removes self intersections, thus overcoming the traditional limitations of mesh-based approaches. To illustrate the effectiveness of TransforMesh, we describe two challenging applications, namely surface morphing and 3-D reconstruction.

Radu Horaud, Florence Forbes, Manuel Yguel et al.

This paper addresses the issue of matching rigid and articulated shapes through probabilistic point registration. The problem is recast into a missing data framework where unknown correspondences are handled via mixture models. Adopting a maximum likelihood principle, we introduce an innovative EM-like algorithm, namely the Expectation Conditional Maximization for Point Registration (ECMPR) algorithm. The algorithm allows the use of general covariance matrices for the mixture model components and improves over the isotropic covariance case. We analyse in detail the associated consequences in terms of estimation of the registration parameters, and we propose an optimal method for estimating the rotational and translational parameters based on semi-definite positive relaxation. We extend rigid registration to articulated registration. Robustness is ensured by detecting and rejecting outliers through the addition of a uniform component to the Gaussian mixture model at hand. We provide an in-depth analysis of our method and we compare it both theoretically and experimentally with other robust methods for point registration.

Miles Hansard, Radu Horaud

The receptive fields of simple cells in the visual cortex can be understood as linear filters. These filters can be modelled by Gabor functions, or by Gaussian derivatives. Gabor functions can also be combined in an `energy model' of the complex cell response. This paper proposes an alternative model of the complex cell, based on Gaussian derivatives. It is most important to account for the insensitivity of the complex response to small shifts of the image. The new model uses a linear combination of the first few derivative filters, at a single position, to approximate the first derivative filter, at a series of adjacent positions. The maximum response, over all positions, gives a signal that is insensitive to small shifts of the image. This model, unlike previous approaches, is based on the scale space theory of visual processing. In particular, the complex cell is built from filters that respond to the \twod\ differential structure of the image. The computational aspects of the new model are studied in one and two dimensions, using the steerability of the Gaussian derivatives. The response of the model to basic images, such as edges and gratings, is derived formally. The response to natural images is also evaluated, using statistical measures of shift insensitivity. The relevance of the new model to the cortical image representation is discussed.

Vasil Khalidov, Florence Forbes, Radu Horaud

The problem of multimodal clustering arises whenever the data are gathered with several physically different sensors. Observations from different modalities are not necessarily aligned in the sense there there is no obvious way to associate or to compare them in some common space. A solution may consist in considering multiple clustering tasks independently for each modality. The main difficulty with such an approach is to guarantee that the unimodal clusterings are mutually consistent. In this paper we show that multimodal clustering can be addressed within a novel framework, namely conjugate mixture models. These models exploit the explicit transformations that are often available between an unobserved parameter space (objects) and each one of the observation spaces (sensors). We formulate the problem as a likelihood maximization task and we derive the associated conjugate expectation-maximization algorithm. The convergence properties of the proposed algorithm are thoroughly investigated. Several local/global optimization techniques are proposed in order to increase its convergence speed. Two initialization strategies are proposed and compared. A consistent model-selection criterion is proposed. The algorithm and its variants are tested and evaluated within the task of 3D localization of several speakers using both auditory and visual data.

Radu Horaud, Matti Niskanen, Guillaume Dewaele et al.

We address the problem of human motion tracking by registering a surface to 3-D data. We propose a method that iteratively computes two things: Maximum likelihood estimates for both the kinematic and free-motion parameters of a kinematic human-body representation, as well as probabilities that the data are assigned either to a body part, or to an outlier cluster. We introduce a new metric between observed points and normals on one side, and a parameterized surface on the other side, the latter being defined as a blending over a set of ellipsoids. We claim that this metric is well suited when one deals with either visual-hull or visual-shape observations. We illustrate the method by tracking human motions using sparse visual-shape data (3-D surface points and normals) gathered from imperfect silhouettes.

Xiaofei Li, Laurent Girin, Fabien Badeig et al.

This paper addresses the problem of sound-source localization (SSL) with a robot head, which remains a challenge in real-world environments. In particular we are interested in locating speech sources, as they are of high interest for human-robot interaction. The microphone-pair response corresponding to the direct-path sound propagation is a function of the source direction. In practice, this response is contaminated by noise and reverberations. The direct-path relative transfer function (DP-RTF) is defined as the ratio between the direct-path acoustic transfer function (ATF) of the two microphones, and it is an important feature for SSL. We propose a method to estimate the DP-RTF from noisy and reverberant signals in the short-time Fourier transform (STFT) domain. First, the convolutive transfer function (CTF) approximation is adopted to accurately represent the impulse response of the microphone array, and the first coefficient of the CTF is mainly composed of the direct-path ATF. At each frequency, the frame-wise speech auto- and cross-power spectral density (PSD) are obtained by spectral subtraction. Then a set of linear equations is constructed by the speech auto- and cross-PSD of multiple frames, in which the DP-RTF is an unknown variable, and is estimated by solving the equations. Finally, the estimated DP-RTFs are concatenated across frequencies and used as a feature vector for SSL. Experiments with a robot, placed in various reverberant environments, show that the proposed method outperforms two state-of-the-art methods.

Xiang Hao, Xiangdong Su, Radu Horaud et al.

This paper proposes a full-band and sub-band fusion model, named as FullSubNet, for single-channel real-time speech enhancement. Full-band and sub-band refer to the models that input full-band and sub-band noisy spectral feature, output full-band and sub-band speech target, respectively. The sub-band model processes each frequency independently. Its input consists of one frequency and several context frequencies. The output is the prediction of the clean speech target for the corresponding frequency. These two types of models have distinct characteristics. The full-band model can capture the global spectral context and the long-distance cross-band dependencies. However, it lacks the ability to modeling signal stationarity and attending the local spectral pattern. The sub-band model is just the opposite. In our proposed FullSubNet, we connect a pure full-band model and a pure sub-band model sequentially and use practical joint training to integrate these two types of models' advantages. We conducted experiments on the DNS challenge (INTERSPEECH 2020) dataset to evaluate the proposed method. Experimental results show that full-band and sub-band information are complementary, and the FullSubNet can effectively integrate them. Besides, the performance of the FullSubNet also exceeds that of the top-ranked methods in the DNS Challenge (INTERSPEECH 2020).

Zhiqi Kang, Mostafa Sadeghi, Radu Horaud

Face frontalization consists of synthesizing a frontally-viewed face from an arbitrarily-viewed one. The main contribution of this paper is a robust face alignment method that enables pixel-to-pixel warping. The method simultaneously estimates the rigid transformation (scale, rotation, and translation) and the non-rigid deformation between two 3D point sets: a set of 3D landmarks extracted from an arbitrary-viewed face, and a set of 3D landmarks parameterized by a frontally-viewed deformable face model. An important merit of the proposed method is its ability to deal both with noise (small perturbations) and with outliers (large errors). We propose to model inliers and outliers with the generalized Student's t-probability distribution function, a heavy-tailed distribution that is immune to non-Gaussian errors in the data. We describe in detail the associated expectation-maximization (EM) algorithm that alternates between the estimation of (i) the rigid parameters, (ii) the deformation parameters, and (iii) the Student-t distribution parameters. We also propose to use the zero-mean normalized cross-correlation, between a frontalized face and the corresponding ground-truth frontally-viewed face, to evaluate the performance of frontalization. To this end, we use a dataset that contains pairs of profile-viewed and frontally-viewed faces. This evaluation, based on direct image-to-image comparison, stands in contrast with indirect evaluation, based on analyzing the effect of frontalization on face recognition.

Sylvain Guy, Stéphane Lathuilière, Pablo Mesejo et al.

Visual voice activity detection (V-VAD) uses visual features to predict whether a person is speaking or not. V-VAD is useful whenever audio VAD (A-VAD) is inefficient either because the acoustic signal is difficult to analyze or because it is simply missing. We propose two deep architectures for V-VAD, one based on facial landmarks and one based on optical flow. Moreover, available datasets, used for learning and for testing V-VAD, lack content variability. We introduce a novel methodology to automatically create and annotate very large datasets in-the-wild -- WildVVAD -- based on combining A-VAD with face detection and tracking. A thorough empirical evaluation shows the advantage of training the proposed deep V-VAD models with this dataset.

Xavier Alameda-Pineda, Vincent Drouard, Radu Horaud

Modeling the temporal behavior of data is of primordial importance in many scientific and engineering fields. Baseline methods assume that both the dynamic and observation equations follow linear-Gaussian models. However, there are many real-world processes that cannot be characterized by a single linear behavior. Alternatively, it is possible to consider a piecewise-linear model which, combined with a switching mechanism, is well suited when several modes of behavior are needed. Nevertheless, switching dynamical systems are intractable because of their computational complexity increases exponentially with time. In this paper, we propose a variational approximation of piecewise linear dynamical systems. We provide full details of the derivation of two variational expectation-maximization algorithms, a filter and a smoother. We show that the model parameters can be split into two sets, static and dynamic parameters, and that the former parameters can be estimated off-line together with the number of linear modes, or the number of states of the switching variable. We apply the proposed method to a visual tracking problem, namely head-pose tracking, and we thoroughly compare our algorithm with several state of the art trackers.

Xiaofei Li, Radu Horaud

This paper proposes a delayed subband LSTM network for online monaural (single-channel) speech enhancement. The proposed method is developed in the short time Fourier transform (STFT) domain. Online processing requires frame-by-frame signal reception and processing. A paramount feature of the proposed method is that the same LSTM is used across frequencies, which drastically reduces the number of network parameters, the amount of training data and the computational burden. Training is performed in a subband manner: the input consists of one frequency, together with a few context frequencies. The network learns a speech-to-noise discriminative function relying on the signal stationarity and on the local spectral pattern, based on which it predicts a clean-speech mask at each frequency. To exploit future information, i.e. look-ahead, we propose an output-delayed subband architecture, which allows the unidirectional forward network to process a few future frames in addition to the current frame. We leverage the proposed method to participate to the DNS real-time speech enhancement challenge. Experiments with the DNS dataset show that the proposed method achieves better performance-measuring scores than the DNS baseline method, which learns the full-band spectra using a gated recurrent unit network.

Mostafa Sadeghi, Xavier Alameda-Pineda, Radu Horaud

This paper addresses the problem of analysing the performance of 3D face alignment (3DFA), or facial landmark localization. This task is usually supervised, based on annotated datasets. Nevertheless, in the particular case of 3DFA, the annotation process is rarely error-free, which strongly biases the results. Alternatively, unsupervised performance analysis (UPA) is investigated. The core ingredient of the proposed methodology is the robust estimation of the rigid transformation between predicted landmarks and model landmarks. It is shown that the rigid mapping thus computed is affected neither by non-rigid facial deformations, due to variabilities in expression and in identity, nor by landmark localization errors, due to various perturbations. The guiding idea is to apply the estimated rotation, translation and scale to a set of predicted landmarks in order to map them onto a mathematical home for the shape embedded in these landmarks (including possible errors). UPA proceeds as follows: (i) 3D landmarks are extracted from a 2D face using the 3DFA method under investigation; (ii) these landmarks are rigidly mapped onto a canonical (frontal) pose, and (iii) a statistically-robust confidence score is computed for each landmark. This allows to assess whether the mapped landmarks lie inside (inliers) or outside (outliers) a confidence volume. An experimental evaluation protocol, that uses publicly available datasets and several 3DFA software packages associated with published articles, is described in detail. The results show that the proposed analysis is consistent with supervised metrics and that it can be used to measure the accuracy of both predicted landmarks and of automatically annotated 3DFA datasets, to detect errors and to eliminate them. Source code and supplemental materials for this paper are publicly available at https://team.inria.fr/robotlearn/upa3dfa/.

Xiaofei LI, Radu Horaud

In this paper, we address the problem of multichannel speech enhancement in the short-time Fourier transform (STFT) domain. A long short-time memory (LSTM) network takes as input a sequence of STFT coefficients associated with a frequency bin of multichannel noisy-speech signals. The network's output is the corresponding sequence of single-channel cleaned speech. We propose several clean-speech network targets, namely, the magnitude ratio mask, the complex STFT coefficients and the (smoothed) spatial filter. A prominent feature of the proposed model is that the same LSTM architecture, with identical parameters, is trained across frequency bins. The proposed method is referred to as narrow-band deep filtering. This choice stays in contrast with traditional wide-band speech enhancement methods. The proposed deep filtering is able to discriminate between speech and noise by exploiting their different temporal and spatial characteristics: speech is non-stationary and spatially coherent while noise is relatively stationary and weakly correlated across channels. This is similar in spirit with unsupervised techniques, such as spectral subtraction and beamforming. We describe extensive experiments with both mixed signals (noise is added to clean speech) and real signals (live recordings). We empirically evaluate the proposed architecture variants using speech enhancement and speech recognition metrics, and we compare our results with the results obtained with several state of the art methods. In the light of these experiments we conclude that narrow-band deep filtering has very good speech enhancement and speech recognition performance, and excellent generalization capabilities in terms of speaker variability and noise type.

Simon Leglaive, Xavier Alameda-Pineda, Laurent Girin et al.

This paper presents a generative approach to speech enhancement based on a recurrent variational autoencoder (RVAE). The deep generative speech model is trained using clean speech signals only, and it is combined with a nonnegative matrix factorization noise model for speech enhancement. We propose a variational expectation-maximization algorithm where the encoder of the RVAE is fine-tuned at test time, to approximate the distribution of the latent variables given the noisy speech observations. Compared with previous approaches based on feed-forward fully-connected architectures, the proposed recurrent deep generative speech model induces a posterior temporal dynamic over the latent variables, which is shown to improve the speech enhancement results.

Mostafa Sadeghi, Simon Leglaive, Xavier Alameda-PIneda et al.

Variational auto-encoders (VAEs) are deep generative latent variable models that can be used for learning the distribution of complex data. VAEs have been successfully used to learn a probabilistic prior over speech signals, which is then used to perform speech enhancement. One advantage of this generative approach is that it does not require pairs of clean and noisy speech signals at training. In this paper, we propose audio-visual variants of VAEs for single-channel and speaker-independent speech enhancement. We develop a conditional VAE (CVAE) where the audio speech generative process is conditioned on visual information of the lip region. At test time, the audio-visual speech generative model is combined with a noise model based on nonnegative matrix factorization, and speech enhancement relies on a Monte Carlo expectation-maximization algorithm. Experiments are conducted with the recently published NTCD-TIMIT dataset as well as the GRID corpus. The results confirm that the proposed audio-visual CVAE effectively fuses audio and visual information, and it improves the speech enhancement performance compared with the audio-only VAE model, especially when the speech signal is highly corrupted by noise. We also show that the proposed unsupervised audio-visual speech enhancement approach outperforms a state-of-the-art supervised deep learning method.

Xiaofei Li, Laurent Girin, Radu Horaud

This paper addresses the problem of under-determinded speech source separation from multichannel microphone singals, i.e. the convolutive mixtures of multiple sources. The time-domain signals are first transformed to the short-time Fourier transform (STFT) domain. To represent the room filters in the STFT domain, instead of the widely-used narrowband assumption, we propose to use a more accurate model, i.e. the convolutive transfer function (CTF). At each frequency band, the CTF coefficients of the mixing filters and the STFT coefficients of the sources are jointly estimated by maximizing the likelihood of the microphone signals, which is resolved by an Expectation-Maximization (EM) algorithm. Experiments show that the proposed method provides very satisfactory performance under highly reverberant environments.

Xiaofei Li, Simon Leglaive, Laurent Girin et al.

We propose a method using a long short-term memory (LSTM) network to estimate the noise power spectral density (PSD) of single-channel audio signals represented in the short time Fourier transform (STFT) domain. An LSTM network common to all frequency bands is trained, which processes each frequency band individually by mapping the noisy STFT magnitude sequence to its corresponding noise PSD sequence. Unlike deep-learning-based speech enhancement methods that learn the full-band spectral structure of speech segments, the proposed method exploits the sub-band STFT magnitude evolution of noise with a long time dependency, in the spirit of the unsupervised noise estimators described in the literature. Speaker- and speech-independent experiments with different types of noise show that the proposed method outperforms the unsupervised estimators, and generalizes well to noise types that are not present in the training set.

Guillaume Delorme, Yihong Xu, Stephane Lathuilière et al.

Unsupervised person re-ID is the task of identifying people on a target data set for which the ID labels are unavailable during training. In this paper, we propose to unify two trends in unsupervised person re-ID: clustering & fine-tuning and adversarial learning. On one side, clustering groups training images into pseudo-ID labels, and uses them to fine-tune the feature extractor. On the other side, adversarial learning is used, inspired by domain adaptation, to match distributions from different domains. Since target data is distributed across different camera viewpoints, we propose to model each camera as an independent domain, and aim to learn domain-independent features. Straightforward adversarial learning yields negative transfer, we thus introduce a conditioning vector to mitigate this undesirable effect. In our framework, the centroid of the cluster to which the visual sample belongs is used as conditioning vector of our conditional adversarial network, where the vector is permutation invariant (clusters ordering does not matter) and its size is independent of the number of clusters. To our knowledge, we are the first to propose the use of conditional adversarial networks for unsupervised person re-ID. We evaluate the proposed architecture on top of two state-of-the-art clustering-based unsupervised person re-identification (re-ID) methods on four different experimental settings with three different data sets and set the new state-of-the-art performance on all four of them. Our code and model will be made publicly available at https://team.inria.fr/perception/canu-reid/.

Benoit Massé, Stéphane Lathuilière, Pablo Mesejo et al.

In this paper we address the problems of detecting objects of interest in a video and of estimating their locations, solely from the gaze directions of people present in the video. Objects can be indistinctly located inside or outside the camera field of view. We refer to this problem as extended gaze following. The contributions of the paper are the followings. First, we propose a novel spatial representation of the gaze directions adopting a top-view perspective. Second, we develop several convolutional encoder/decoder networks to predict object locations and compare them with heuristics and with classical learning-based approaches. Third, in order to train the proposed models, we generate a very large number of synthetic scenarios employing a probabilistic formulation. Finally, our methodology is empirically validated using a publicly available dataset.

Simon Leglaive, Umut Simsekli, Antoine Liutkus et al.

This paper focuses on single-channel semi-supervised speech enhancement. We learn a speaker-independent deep generative speech model using the framework of variational autoencoders. The noise model remains unsupervised because we do not assume prior knowledge of the noisy recording environment. In this context, our contribution is to propose a noise model based on alpha-stable distributions, instead of the more conventional Gaussian non-negative matrix factorization approach found in previous studies. We develop a Monte Carlo expectation-maximization algorithm for estimating the model parameters at test time. Experimental results show the superiority of the proposed approach both in terms of perceptual quality and intelligibility of the enhanced speech signal.

Simon Leglaive, Laurent Girin, Radu Horaud

In this paper we address the problem of enhancing speech signals in noisy mixtures using a source separation approach. We explore the use of neural networks as an alternative to a popular speech variance model based on supervised non-negative matrix factorization (NMF). More precisely, we use a variational autoencoder as a speaker-independent supervised generative speech model, highlighting the conceptual similarities that this approach shares with its NMF-based counterpart. In order to be free of generalization issues regarding the noisy recording environments, we follow the approach of having a supervised model only for the target speech signal, the noise model being based on unsupervised NMF. We develop a Monte Carlo expectation-maximization algorithm for inferring the latent variables in the variational autoencoder and estimating the unsupervised model parameters. Experiments show that the proposed method outperforms a semi-supervised NMF baseline and a state-of-the-art fully supervised deep learning approach.

Xiaofei Li, Laurent Girin, Sharon Gannot et al.

This paper addresses the problem of multichannel online dereverberation. The proposed method is carried out in the short-time Fourier transform (STFT) domain, and for each frequency band independently. In the STFT domain, the time-domain room impulse response is approximately represented by the convolutive transfer function (CTF). The multichannel CTFs are adaptively identified based on the cross-relation method, and using the recursive least square criterion. Instead of the complex-valued CTF convolution model, we use a nonnegative convolution model between the STFT magnitude of the source signal and the CTF magnitude, which is just a coarse approximation of the former model, but is shown to be more robust against the CTF perturbations. Based on this nonnegative model, we propose an online STFT magnitude inverse filtering method. The inverse filters of the CTF magnitude are formulated based on the multiple-input/output inverse theorem (MINT), and adaptively estimated based on the gradient descent criterion. Finally, the inverse filtering is applied to the STFT magnitude of the microphone signals, obtaining an estimate of the STFT magnitude of the source signal. Experiments regarding both speech enhancement and automatic speech recognition are conducted, which demonstrate that the proposed method can effectively suppress reverberation, even for the difficult case of a moving speaker.

Yutong Ban, Xavier Alameda-PIneda, Christine Evers et al.

In this paper we address the problem of simultaneously tracking several moving audio sources, namely the problem of estimating source trajectories from a sequence of observed features. We propose to use the von Mises distribution to model audio-source directions of arrival with circular random variables. This leads to a Bayesian filtering formulation which is intractable because of the combinatorial explosion of associating observed variables with latent variables, over time. We propose a variational approximation of the filtering distribution. We infer a variational expectation-maximization algorithm that is both computationally tractable and time efficient. We propose an audio-source birth method that favors smooth source trajectories and which is used both to initialize the number of active sources and to detect new sources. We perform experiments with the recently released LOCATA dataset comprising two moving sources and a moving microphone array mounted onto a robot.

Xiaofei Li, Yutong Ban, Laurent Girin et al.

This paper presents an online multiple-speaker localization and tracking method, as the INRIA-Perception contribution to the LOCATA Challenge 2018. First, the recursive least-square method is used to adaptively estimate the direct-path relative transfer function as an interchannel localization feature. The feature is assumed to associate with a single speaker at each time-frequency bin. Second, a complex Gaussian mixture model (CGMM) is used as a generative model of the features. The weight of each CGMM component represents the probability that this component corresponds to an active speaker, and is adaptively estimated with an online optimization algorithm. Finally, taking the CGMM component weights as observations, a Bayesian multiple-speaker tracking method based on the variational expectation maximization algorithm is used. The tracker accounts for the variation of active speakers and the localization miss measurements, by introducing speaker birth and sleeping processes. The experiments carried out on the development dataset of the challenge are reported.

Simon Leglaive, Laurent Girin, Radu Horaud

In this paper we address speaker-independent multichannel speech enhancement in unknown noisy environments. Our work is based on a well-established multichannel local Gaussian modeling framework. We propose to use a neural network for modeling the speech spectro-temporal content. The parameters of this supervised model are learned using the framework of variational autoencoders. The noisy recording environment is supposed to be unknown, so the noise spectro-temporal modeling remains unsupervised and is based on non-negative matrix factorization (NMF). We develop a Monte Carlo expectation-maximization algorithm and we experimentally show that the proposed approach outperforms its NMF-based counterpart, where speech is modeled using supervised NMF.

Yutong Ban, Xavier Alameda-Pineda, Laurent Girin et al.

In this paper we address the problem of tracking multiple speakers via the fusion of visual and auditory information. We propose to exploit the complementary nature of these two modalities in order to accurately estimate smooth trajectories of the tracked persons, to deal with the partial or total absence of one of the modalities over short periods of time, and to estimate the acoustic status -- either speaking or silent -- of each tracked person along time. We propose to cast the problem at hand into a generative audio-visual fusion (or association) model formulated as a latent-variable temporal graphical model. This may well be viewed as the problem of maximizing the posterior joint distribution of a set of continuous and discrete latent variables given the past and current observations, which is intractable. We propose a variational inference model which amounts to approximate the joint distribution with a factorized distribution. The solution takes the form of a closed-form expectation maximization procedure. We describe in detail the inference algorithm, we evaluate its performance and we compare it with several baseline methods. These experiments show that the proposed audio-visual tracker performs well in informal meetings involving a time-varying number of people.

Xiaofei Li, Yutong Ban, Laurent Girin et al.

We address the problem of online localization and tracking of multiple moving speakers in reverberant environments. The paper has the following contributions. We use the direct-path relative transfer function (DP-RTF), an inter-channel feature that encodes acoustic information robust against reverberation, and we propose an online algorithm well suited for estimating DP-RTFs associated with moving audio sources. Another crucial ingredient of the proposed method is its ability to properly assign DP-RTFs to audio-source directions. Towards this goal, we adopt a maximum-likelihood formulation and we propose to use an exponentiated gradient (EG) to efficiently update source-direction estimates starting from their currently available values. The problem of multiple speaker tracking is computationally intractable because the number of possible associations between observed source directions and physical speakers grows exponentially with time. We adopt a Bayesian framework and we propose a variational approximation of the posterior filtering distribution associated with multiple speaker tracking, as well as an efficient variational expectation-maximization (VEM) solver. The proposed online localization and tracking method is thoroughly evaluated using two datasets that contain recordings performed in real environments.

Stéphane Lathuilière, Pablo Mesejo, Xavier Alameda-Pineda et al.

In this paper, we address the problem of how to robustly train a ConvNet for regression, or deep robust regression. Traditionally, deep regression employs the L2 loss function, known to be sensitive to outliers, i.e. samples that either lie at an abnormal distance away from the majority of the training samples, or that correspond to wrongly annotated targets. This means that, during back-propagation, outliers may bias the training process due to the high magnitude of their gradient. In this paper, we propose DeepGUM: a deep regression model that is robust to outliers thanks to the use of a Gaussian-uniform mixture model. We derive an optimization algorithm that alternates between the unsupervised detection of outliers using expectation-maximization, and the supervised training with cleaned samples using stochastic gradient descent. DeepGUM is able to adapt to a continuously evolving outlier distribution, avoiding to manually impose any threshold on the proportion of outliers in the training set. Extensive experimental evaluations on four different tasks (facial and fashion landmark detection, age and head pose estimation) lead us to conclude that our novel robust technique provides reliability in the presence of various types of noise and protection against a high percentage of outliers.

Stéphane Lathuilière, Pablo Mesejo, Xavier Alameda-Pineda et al.

Deep learning revolutionized data science, and recently its popularity has grown exponentially, as did the amount of papers employing deep networks. Vision tasks, such as human pose estimation, did not escape from this trend. There is a large number of deep models, where small changes in the network architecture, or in the data pre-processing, together with the stochastic nature of the optimization procedures, produce notably different results, making extremely difficult to sift methods that significantly outperform others. This situation motivates the current study, in which we perform a systematic evaluation and statistical analysis of vanilla deep regression, i.e. convolutional neural networks with a linear regression top layer. This is the first comprehensive analysis of deep regression techniques. We perform experiments on four vision problems, and report confidence intervals for the median performance as well as the statistical significance of the results, if any. Surprisingly, the variability due to different data pre-processing procedures generally eclipses the variability due to modifications in the network architecture. Our results reinforce the hypothesis according to which, in general, a general-purpose network (e.g. VGG-16 or ResNet-50) adequately tuned can yield results close to the state-of-the-art without having to resort to more complex and ad-hoc regression models.

Xiaofei Li, Laurent Girin, Sharon Gannot et al.

This paper addresses the problem of speech separation and enhancement from multichannel convolutive and noisy mixtures, \emph{assuming known mixing filters}. We propose to perform the speech separation and enhancement task in the short-time Fourier transform domain, using the convolutive transfer function (CTF) approximation. Compared to time-domain filters, CTF has much less taps, consequently it has less near-common zeros among channels and less computational complexity. The work proposes three speech-source recovery methods, namely: i) the multichannel inverse filtering method, i.e. the multiple input/output inverse theorem (MINT), is exploited in the CTF domain, and for the multi-source case, ii) a beamforming-like multichannel inverse filtering method applying single source MINT and using power minimization, which is suitable whenever the source CTFs are not all known, and iii) a constrained Lasso method, where the sources are recovered by minimizing the $\ell_1$-norm to impose their spectral sparsity, with the constraint that the $\ell_2$-norm fitting cost, between the microphone signals and the mixing model involving the unknown source signals, is less than a tolerance. The noise can be reduced by setting a tolerance onto the noise power. Experiments under various acoustic conditions are carried out to evaluate the three proposed methods. The comparison between them as well as with the baseline methods is presented.

Stéphane Lathuilière, Benoit Massé, Pablo Mesejo et al.

This paper introduces a novel neural network-based reinforcement learning approach for robot gaze control. Our approach enables a robot to learn and to adapt its gaze control strategy for human-robot interaction neither with the use of external sensors nor with human supervision. The robot learns to focus its attention onto groups of people from its own audio-visual experiences, independently of the number of people, of their positions and of their physical appearances. In particular, we use a recurrent neural network architecture in combination with Q-learning to find an optimal action-selection policy; we pre-train the network using a simulated environment that mimics realistic scenarios that involve speaking/silent participants, thus avoiding the need of tedious sessions of a robot interacting with people. Our experimental evaluation suggests that the proposed method is robust against parameter estimation, i.e. the parameter values yielded by the method do not have a decisive impact on the performance. The best results are obtained when both audio and visual information is jointly used. Experiments with the Nao robot indicate that our framework is a step forward towards the autonomous learning of socially acceptable gaze behavior.

Richard T. Marriott, Alexander Paschevich, Radu Horaud

We propose a novel algorithm for unsupervised extraction of piecewise planar models from depth-data. Among other applications, such models are a good way of enabling autonomous agents (robots, cars, drones, etc.) to effectively perceive their surroundings and to navigate in three dimensions. We propose to do this by fitting the data with a piecewise-linear Gaussian mixture regression model whose components are skewed over planes, making them flat in appearance rather than being ellipsoidal, by embedding an outlier-trimming process that is formally incorporated into the proposed expectation-maximization algorithm, and by selectively fusing contiguous, coplanar components. Part of our motivation is an attempt to estimate more accurate plane-extraction by allowing each model component to make use of all available data through probabilistic clustering. The algorithm is thoroughly evaluated against a standard benchmark and is shown to rank among the best of the existing state-of-the-art methods.

Xiaofei Li, Radu Horaud, Sharon Gannot

This paper addresses the problems of blind channel identification and multichannel equalization for speech dereverberation and noise reduction. The time-domain cross-relation method is not suitable for blind room impulse response identification, due to the near-common zeros of the long impulse responses. We extend the cross-relation method to the short-time Fourier transform (STFT) domain, in which the time-domain impulse responses are approximately represented by the convolutive transfer functions (CTFs) with much less coefficients. The CTFs suffer from the common zeros caused by the oversampled STFT. We propose to identify CTFs based on the STFT with the oversampled signals and the critical sampled CTFs, which is a good compromise between the frequency aliasing of the signals and the common zeros problem of CTFs. In addition, a normalization of the CTFs is proposed to remove the gain ambiguity across sub-bands. In the STFT domain, the identified CTFs is used for multichannel equalization, in which the sparsity of speech signals is exploited. We propose to perform inverse filtering by minimizing the $\ell_1$-norm of the source signal with the relaxed $\ell_2$-norm fitting error between the micophone signals and the convolution of the estimated source signal and the CTFs used as a constraint. This method is advantageous in that the noise can be reduced by relaxing the $\ell_2$-norm to a tolerance corresponding to the noise power, and the tolerance can be automatically set. The experiments confirm the efficiency of the proposed method even under conditions with high reverberation levels and intense noise.

Benoît Massé, Silèye Ba, Radu Horaud

The visual focus of attention (VFOA) has been recognized as a prominent conversational cue. We are interested in estimating and tracking the VFOAs associated with multi-party social interactions. We note that in this type of situations the participants either look at each other or at an object of interest; therefore their eyes are not always visible. Consequently both gaze and VFOA estimation cannot be based on eye detection and tracking. We propose a method that exploits the correlation between eye gaze and head movements. Both VFOA and gaze are modeled as latent variables in a Bayesian switching state-space model. The proposed formulation leads to a tractable learning procedure and to an efficient algorithm that simultaneously tracks gaze and visual focus. The method is tested and benchmarked using two publicly available datasets that contain typical multi-party human-robot and human-human interactions.