Xavier Alameda-Pineda

Wen Guo, Yuming Du, Xi Shen et al. · tencent-ai

This paper tackles the problem of human motion prediction, consisting in forecasting future body poses from historically observed sequences. State-of-the-art approaches provide good results, however, they rely on deep learning architectures of arbitrary complexity, such as Recurrent Neural Networks(RNN), Transformers or Graph Convolutional Networks(GCN), typically requiring multiple training stages and more than 2 million parameters. In this paper, we show that, after combining with a series of standard practices, such as applying Discrete Cosine Transform(DCT), predicting residual displacement of joints and optimizing velocity as an auxiliary loss, a light-weight network based on multi-layer perceptrons(MLPs) with only 0.14 million parameters can surpass the state-of-the-art performance. An exhaustive evaluation on the Human3.6M, AMASS, and 3DPW datasets shows that our method, named siMLPe, consistently outperforms all other approaches. We hope that our simple method could serve as a strong baseline for the community and allow re-thinking of the human motion prediction problem. The code is publicly available at \url{https://github.com/dulucas/siMLPe}.

Guanglei Yang, Enrico Fini, Dan Xu et al.

A fundamental and challenging problem in deep learning is catastrophic forgetting, i.e. the tendency of neural networks to fail to preserve the knowledge acquired from old tasks when learning new tasks. This problem has been widely investigated in the research community and several Incremental Learning (IL) approaches have been proposed in the past years. While earlier works in computer vision have mostly focused on image classification and object detection, more recently some IL approaches for semantic segmentation have been introduced. These previous works showed that, despite its simplicity, knowledge distillation can be effectively employed to alleviate catastrophic forgetting. In this paper, we follow this research direction and, inspired by recent literature on contrastive learning, we propose a novel distillation framework, Uncertainty-aware Contrastive Distillation (\method). In a nutshell, \method~is operated by introducing a novel distillation loss that takes into account all the images in a mini-batch, enforcing similarity between features associated to all the pixels from the same classes, and pulling apart those corresponding to pixels from different classes. In order to mitigate catastrophic forgetting, we contrast features of the new model with features extracted by a frozen model learned at the previous incremental step. Our experimental results demonstrate the advantage of the proposed distillation technique, which can be used in synergy with previous IL approaches, and leads to state-of-art performance on three commonly adopted benchmarks for incremental semantic segmentation. The code is available at \url{https://github.com/ygjwd12345/UCD}.

Jean-Eudes Ayilo, Mostafa Sadeghi, Romain Serizel et al.

This paper addresses unsupervised diffusion-based single-channel speech enhancement (SE). Prior work in this direction combines a score-based diffusion model trained on clean speech with a Gaussian noise model whose covariance is structured by non-negative matrix factorization (NMF). This combination is used within an iterative expectation-maximization (EM) scheme, in which a diffusion-based posterior-sampling E-step estimates the clean speech. We first revisit this framework and propose to explicitly model both speech and acoustic noise as latent variables, jointly sampling them in the E-step instead of sampling speech alone as in previous approaches. We then introduce a new semi-supervised SE framework that replaces the NMF noise prior with a diffusion-based noise model, learned jointly with the speech prior in a single conditional score model. Within this framework, we derive two variants: one that implicitly accounts for noise and one that explicitly treats noise as a latent variable. Experiments on WSJ0-QUT and VoiceBank-DEMAND show that explicit noise modeling systematically improves SE performance for both NMF-based and diffusion-based noise priors. Under matched conditions, the diffusion-based noise model attains the best overall quality and intelligibility among unsupervised methods, while under mismatched conditions the proposed NMF-based explicit-noise framework is more robust and suffers less degradation than several supervised baselines. Code, demo, and supplementary materials are publicly available.

Enrico Fini, Pietro Astolfi, Karteek Alahari et al.

Self-supervised learning models have been shown to learn rich visual representations without requiring human annotations. However, in many real-world scenarios, labels are partially available, motivating a recent line of work on semi-supervised methods inspired by self-supervised principles. In this paper, we propose a conceptually simple yet empirically powerful approach to turn clustering-based self-supervised methods such as SwAV or DINO into semi-supervised learners. More precisely, we introduce a multi-task framework merging a supervised objective using ground-truth labels and a self-supervised objective relying on clustering assignments with a single cross-entropy loss. This approach may be interpreted as imposing the cluster centroids to be class prototypes. Despite its simplicity, we provide empirical evidence that our approach is highly effective and achieves state-of-the-art performance on CIFAR100 and ImageNet.

Victor Sanchez, Chris Reinke, Ahamed Mohamed et al.

In this paper, we introduce OpenSocInt, an open-source software package providing a simulator for multi-modal social interactions and a modular architecture to train social agents. We described the software package and showcased its interest via an experimental protocol based on the task of social navigation. Our framework allows for exploring the use of different perceptual features, their encoding and fusion, as well as the use of different agents. The software is already publicly available under GPL at https://gitlab.inria.fr/robotlearn/OpenSocInt/.

Thomas De Min, Massimiliano Mancini, Karteek Alahari et al.

State-of-the-art rehearsal-free continual learning methods exploit the peculiarities of Vision Transformers to learn task-specific prompts, drastically reducing catastrophic forgetting. However, there is a tradeoff between the number of learned parameters and the performance, making such models computationally expensive. In this work, we aim to reduce this cost while maintaining competitive performance. We achieve this by revisiting and extending a simple transfer learning idea: learning task-specific normalization layers. Specifically, we tune the scale and bias parameters of LayerNorm for each continual learning task, selecting them at inference time based on the similarity between task-specific keys and the output of the pre-trained model. To make the classifier robust to incorrect selection of parameters during inference, we introduce a two-stage training procedure, where we first optimize the task-specific parameters and then train the classifier with the same selection procedure of the inference time. Experiments on ImageNet-R and CIFAR-100 show that our method achieves results that are either superior or on par with {the state of the art} while being computationally cheaper.

Xiaoyu Bie, Wen Guo, Simon Leglaive et al.

Studies on the automatic processing of 3D human pose data have flourished in the recent past. In this paper, we are interested in the generation of plausible and diverse future human poses following an observed 3D pose sequence. Current methods address this problem by injecting random variables from a single latent space into a deterministic motion prediction framework, which precludes the inherent multi-modality in human motion generation. In addition, previous works rarely explore the use of attention to select which frames are to be used to inform the generation process up to our knowledge. To overcome these limitations, we propose Hierarchical Transformer Dynamical Variational Autoencoder, HiT-DVAE, which implements auto-regressive generation with transformer-like attention mechanisms. HiT-DVAE simultaneously learns the evolution of data and latent space distribution with time correlated probabilistic dependencies, thus enabling the generative model to learn a more complex and time-varying latent space as well as diverse and realistic human motions. Furthermore, the auto-regressive generation brings more flexibility on observation and prediction, i.e. one can have any length of observation and predict arbitrary large sequences of poses with a single pre-trained model. We evaluate the proposed method on HumanEva-I and Human3.6M with various evaluation methods, and outperform the state-of-the-art methods on most of the metrics.

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/

Anand Ballou, Xavier Alameda-Pineda, Chris Reinke

With the increasing presence of robots in our every-day environments, improving their social skills is of utmost importance. Nonetheless, social robotics still faces many challenges. One bottleneck is that robotic behaviors need to be often adapted as social norms depend strongly on the environment. For example, a robot should navigate more carefully around patients in a hospital compared to workers in an office. In this work, we investigate meta-reinforcement learning (meta-RL) as a potential solution. Here, robot behaviors are learned via reinforcement learning where a reward function needs to be chosen so that the robot learns an appropriate behavior for a given environment. We propose to use a variational meta-RL procedure that quickly adapts the robots' behavior to new reward functions. As a result, given a new environment different reward functions can be quickly evaluated and an appropriate one selected. The procedure learns a vectorized representation for reward functions and a meta-policy that can be conditioned on such a representation. Given observations from a new reward function, the procedure identifies its representation and conditions the meta-policy to it. While investigating the procedures' capabilities, we realized that it suffers from posterior collapse where only a subset of the dimensions in the representation encode useful information resulting in a reduced performance. Our second contribution, a radial basis function (RBF) layer, partially mitigates this negative effect. The RBF layer lifts the representation to a higher dimensional space, which is more easily exploitable for the meta-policy. We demonstrate the interest of the RBF layer and the usage of meta-RL for social robotics on four robotic simulation tasks.

Samir Sadok, Simon Leglaive, Laurent Girin et al.

Understanding and controlling latent representations in deep generative models is a challenging yet important problem for analyzing, transforming and generating various types of data. In speech processing, inspiring from the anatomical mechanisms of phonation, the source-filter model considers that speech signals are produced from a few independent and physically meaningful continuous latent factors, among which the fundamental frequency $f_0$ and the formants are of primary importance. In this work, we start from a variational autoencoder (VAE) trained in an unsupervised manner on a large dataset of unlabeled natural speech signals, and we show that the source-filter model of speech production naturally arises as orthogonal subspaces of the VAE latent space. Using only a few seconds of labeled speech signals generated with an artificial speech synthesizer, we propose a method to identify the latent subspaces encoding $f_0$ and the first three formant frequencies, we show that these subspaces are orthogonal, and based on this orthogonality, we develop a method to accurately and independently control the source-filter speech factors within the latent subspaces. Without requiring additional information such as text or human-labeled data, this results in a deep generative model of speech spectrograms that is conditioned on $f_0$ and the formant frequencies, and which is applied to the transformation speech signals. Finally, we also propose a robust $f_0$ estimation method that exploits the projection of a speech signal onto the learned latent subspace associated with $f_0$.

Xiaoyu Lin, Simon Leglaive, Laurent Girin et al.

This work builds on a previous work on unsupervised speech enhancement using a dynamical variational autoencoder (DVAE) as the clean speech model and non-negative matrix factorization (NMF) as the noise model. We propose to replace the NMF noise model with a deep dynamical generative model (DDGM) depending either on the DVAE latent variables, or on the noisy observations, or on both. This DDGM can be trained in three configurations: noise-agnostic, noise-dependent and noise adaptation after noise-dependent training. Experimental results show that the proposed method achieves competitive performance compared to state-of-the-art unsupervised speech enhancement methods, while the noise-dependent training configuration yields a much more time-efficient inference process.

Xiaoyu Lin, Xiaoyu Bie, Simon Leglaive et al.

The dynamical variational autoencoders (DVAEs) are a family of latent-variable deep generative models that extends the VAE to model a sequence of observed data and a corresponding sequence of latent vectors. In almost all the DVAEs of the literature, the temporal dependencies within each sequence and across the two sequences are modeled with recurrent neural networks. In this paper, we propose to model speech signals with the Hierarchical Transformer DVAE (HiT-DVAE), which is a DVAE with two levels of latent variable (sequence-wise and frame-wise) and in which the temporal dependencies are implemented with the Transformer architecture. We show that HiT-DVAE outperforms several other DVAEs for speech spectrogram modeling, while enabling a simpler training procedure, revealing its high potential for downstream low-level speech processing tasks such as speech enhancement.

Louis Airale, Xavier Alameda-Pineda, Stéphane Lathuilière et al.

In this work, we address the task of unconditional head motion generation to animate still human faces in a low-dimensional semantic space from a single reference pose. Different from traditional audio-conditioned talking head generation that seldom puts emphasis on realistic head motions, we devise a GAN-based architecture that learns to synthesize rich head motion sequences over long duration while maintaining low error accumulation levels.In particular, the autoregressive generation of incremental outputs ensures smooth trajectories, while a multi-scale discriminator on input pairs drives generation toward better handling of high- and low-frequency signals and less mode collapse.We experimentally demonstrate the relevance of the proposed method and show its superiority compared to models that attained state-of-the-art performances on similar tasks.

Guénolé Fiche, Simon Leglaive, Xavier Alameda-Pineda et al.

Pose and motion priors are crucial for recovering realistic and accurate human motion from noisy observations. Substantial progress has been made on pose and shape estimation from images, and recent works showed impressive results using priors to refine frame-wise predictions. However, a lot of motion priors only model transitions between consecutive poses and are used in time-consuming optimization procedures, which is problematic for many applications requiring real-time motion capture. We introduce Motion-DVAE, a motion prior to capture the short-term dependencies of human motion. As part of the dynamical variational autoencoder (DVAE) models family, Motion-DVAE combines the generative capability of VAE models and the temporal modeling of recurrent architectures. Together with Motion-DVAE, we introduce an unsupervised learned denoising method unifying regression- and optimization-based approaches in a single framework for real-time 3D human pose estimation. Experiments show that the proposed approach reaches competitive performance with state-of-the-art methods while being much faster.

Ali Golmakani, Mostafa Sadeghi, Xavier Alameda-Pineda et al.

We address speech enhancement based on variational autoencoders, which involves learning a speech prior distribution in the time-frequency (TF) domain. A zero-mean complex-valued Gaussian distribution is usually assumed for the generative model, where the speech information is encoded in the variance as a function of a latent variable. In contrast to this commonly used approach, we propose a weighted variance generative model, where the contribution of each spectrogram time-frame in parameter learning is weighted. We impose a Gamma prior distribution on the weights, which would effectively lead to a Student's t-distribution instead of Gaussian for speech generative modeling. We develop efficient training and speech enhancement algorithms based on the proposed generative model. Our experimental results on spectrogram auto-encoding and speech enhancement demonstrate the effectiveness and robustness of the proposed approach compared to the standard unweighted variance model.

Louis Airale, Dominique Vaufreydaz, Xavier Alameda-Pineda

Animating still face images with deep generative models using a speech input signal is an active research topic and has seen important recent progress.However, much of the effort has been put into lip syncing and rendering quality while the generation of natural head motion, let alone the audio-visual correlation between head motion and speech, has often been neglected.In this work, we propose a multi-scale audio-visual synchrony loss and a multi-scale autoregressive GAN to better handle short and long-term correlation between speech and the dynamics of the head and lips.In particular, we train a stack of syncer models on multimodal input pyramids and use these models as guidance in a multi-scale generator network to produce audio-aligned motion unfolding over diverse time scales.Both the pyramid of audio-visual syncers and the generative models are trained in a low-dimensional space that fully preserves dynamics cues.The experiments show significant improvements over the state-of-the-art in head motion dynamics quality and especially in multi-scale audio-visual synchrony on a collection of benchmark datasets.

Samir Sadok, Stéphane Lathuilière, Xavier Alameda-Pineda

Recent speech modeling relies on explicit attributes such as pitch, content, and speaker identity, but these alone cannot capture the full richness of natural speech. We introduce RT-MAE, a novel masked autoencoder framework that augments the supervised attributes-based modeling with unsupervised residual trainable tokens, designed to encode the information not explained by explicit labeled factors (e.g., timbre variations, noise, emotion etc). Experiments show that RT-MAE improves reconstruction quality, preserving content and speaker similarity while enhancing expressivity. We further demonstrate its applicability to speech enhancement, removing noise at inference while maintaining controllability and naturalness.

Samir Sadok, Laurent Girin, Xavier Alameda-Pineda

Neural audio codecs (NACs) typically encode the short-term energy (gain) and normalized structure (shape) of speech/audio signals jointly within the same latent space. As a result, they are poorly robust to a global variation of the input signal level in the sense that such variation has strong influence on the embedding vectors at the output of the encoder and their quantization. This methodology is inherently inefficient, leading to codebook redundancy and suboptimal bitrate-distortion performance. To address these limitations, we propose to introduce shape-gain decomposition, widely used in classical speech/audio coding, into the NAC framework. The principle of the proposed Equalizer methodology is to decompose the input signal -- before the NAC encoder -- into gain and normalized shape vector on a short-term basis. The shape vector is processed by the NAC, while the gain is quantized with scalar quantization and transmitted separately. The output (decoded) signal is reconstructed from the normalized output of the NAC and the quantized gain. Our experiments conducted on speech signals show that this general methodology, easily applicable to any NAC, enables a substantial gain in bitrate-distortion performance, as well as a massive reduction in complexity.

Daniel Jost, Basavasagar Patil, Xavier Alameda-Pineda et al.

Deep Neural Networks (DNNs) became the standard tool for function approximation with most of the introduced architectures being developed for high-dimensional input data. However, many real-world problems have low-dimensional inputs for which standard Multi-Layer Perceptrons (MLPs) are the default choice. An investigation into specialized architectures is missing. We propose a novel DNN layer called Univariate Radial Basis Function (U-RBF) layer as an alternative. Similar to sensory neurons in the brain, the U-RBF layer processes each individual input dimension with a population of neurons whose activations depend on different preferred input values. We verify its effectiveness compared to MLPs in low-dimensional function regressions and reinforcement learning tasks. The results show that the U-RBF is especially advantageous when the target function becomes complex and difficult to approximate.

Massimiliano Pappa, Luca Romani, Valentino Sacco et al.

Deploying safety-critical agents requires anticipating the consequences of actions before they are executed. While world models offer a paradigm for this proactive foresight, current approaches relying on visual simulation incur prohibitive latencies, often exceeding several seconds per step. In this work, we challenge the assumption that visual processing is necessary for failure prevention. We show that a trained policy's latent state, combined with its planned actions, already encodes sufficient information to anticipate action outcomes, making visual simulation redundant for failure prevention. To this end, we introduce DILLO (DIstiLLed Language-ActiOn World Model), a fast steering layer that shifts the paradigm from "simulate-then-act" to "describe-then-act." DILLO is trained via cross-modal distillation, where a privileged Vision Language Model teacher annotates offline trajectories and a latent-conditioned Large Language Model student learns to predict semantic outcomes. This creates a text-only inference path, bypassing heavy visual generation entirely, achieving a 14x speedup over baselines. Experiments on MetaWorld and LIBERO demonstrate that DILLO produces high-fidelity descriptions of the next state and is able to steer the policy, improving episode success rate by up to 15 pp and 9.3 pp on average across tasks.

Lorenzo Vaquero, Yihong Xu, Xavier Alameda-Pineda et al.

Multi-object tracking (MOT) endeavors to precisely estimate the positions and identities of multiple objects over time. The prevailing approach, tracking-by-detection (TbD), first detects objects and then links detections, resulting in a simple yet effective method. However, contemporary detectors may occasionally miss some objects in certain frames, causing trackers to cease tracking prematurely. To tackle this issue, we propose BUSCA, meaning `to search', a versatile framework compatible with any online TbD system, enhancing its ability to persistently track those objects missed by the detector, primarily due to occlusions. Remarkably, this is accomplished without modifying past tracking results or accessing future frames, i.e., in a fully online manner. BUSCA generates proposals based on neighboring tracks, motion, and learned tokens. Utilizing a decision Transformer that integrates multimodal visual and spatiotemporal information, it addresses the object-proposal association as a multi-choice question-answering task. BUSCA is trained independently of the underlying tracker, solely on synthetic data, without requiring fine-tuning. Through BUSCA, we showcase consistent performance enhancements across five different trackers and establish a new state-of-the-art baseline across three different benchmarks. Code available at: https://github.com/lorenzovaquero/BUSCA.

Yihong Xu, Yutong Ban, Guillaume Delorme et al.

Transformers have proven superior performance for a wide variety of tasks since they were introduced. In recent years, they have drawn attention from the vision community in tasks such as image classification and object detection. Despite this wave, an accurate and efficient multiple-object tracking (MOT) method based on transformers is yet to be designed. We argue that the direct application of a transformer architecture with quadratic complexity and insufficient noise-initialized sparse queries - is not optimal for MOT. We propose TransCenter, a transformer-based MOT architecture with dense representations for accurately tracking all the objects while keeping a reasonable runtime. Methodologically, we propose the use of image-related dense detection queries and efficient sparse tracking queries produced by our carefully designed query learning networks (QLN). On one hand, the dense image-related detection queries allow us to infer targets' locations globally and robustly through dense heatmap outputs. On the other hand, the set of sparse tracking queries efficiently interacts with image features in our TransCenter Decoder to associate object positions through time. As a result, TransCenter exhibits remarkable performance improvements and outperforms by a large margin the current state-of-the-art methods in two standard MOT benchmarks with two tracking settings (public/private). TransCenter is also proven efficient and accurate by an extensive ablation study and comparisons to more naive alternatives and concurrent works. For scientific interest, the code is made publicly available at https://github.com/yihongxu/transcenter.

Guanglei Yang, Paolo Rota, Xavier Alameda-Pineda et al.

Convolutional neural networks have enabled major progresses in addressing pixel-level prediction tasks such as semantic segmentation, depth estimation, surface normal prediction and so on, benefiting from their powerful capabilities in visual representation learning. Typically, state of the art models integrate attention mechanisms for improved deep feature representations. Recently, some works have demonstrated the significance of learning and combining both spatial- and channelwise attentions for deep feature refinement. In this paper, weaim at effectively boosting previous approaches and propose a unified deep framework to jointly learn both spatial attention maps and channel attention vectors in a principled manner so as to structure the resulting attention tensors and model interactions between these two types of attentions. Specifically, we integrate the estimation and the interaction of the attentions within a probabilistic representation learning framework, leading to VarIational STructured Attention networks (VISTA-Net). We implement the inference rules within the neural network, thus allowing for end-to-end learning of the probabilistic and the CNN frontend parameters. As demonstrated by our extensive empirical evaluation on six large-scale datasets for dense visual prediction, VISTA-Net outperforms the state-of-the-art in multiple continuous and discrete prediction tasks, thus confirming the benefit of the proposed approach in joint structured spatial-channel attention estimation for deep representation learning. The code is available at https://github.com/ygjwd12345/VISTA-Net.

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.

Guénolé Fiche, Simon Leglaive, Xavier Alameda-Pineda et al.

Previous works on Human Pose and Shape Estimation (HPSE) from RGB images can be broadly categorized into two main groups: parametric and non-parametric approaches. Parametric techniques leverage a low-dimensional statistical body model for realistic results, whereas recent non-parametric methods achieve higher precision by directly regressing the 3D coordinates of the human body mesh. This work introduces a novel paradigm to address the HPSE problem, involving a low-dimensional discrete latent representation of the human mesh and framing HPSE as a classification task. Instead of predicting body model parameters or 3D vertex coordinates, we focus on predicting the proposed discrete latent representation, which can be decoded into a registered human mesh. This innovative paradigm offers two key advantages. Firstly, predicting a low-dimensional discrete representation confines our predictions to the space of anthropomorphic poses and shapes even when little training data is available. Secondly, by framing the problem as a classification task, we can harness the discriminative power inherent in neural networks. The proposed model, VQ-HPS, predicts the discrete latent representation of the mesh. The experimental results demonstrate that VQ-HPS outperforms the current state-of-the-art non-parametric approaches while yielding results as realistic as those produced by parametric methods when trained with little data. VQ-HPS also shows promising results when training on large-scale datasets, highlighting the significant potential of the classification approach for HPSE. See the project page at https://g-fiche.github.io/research-pages/vqhps/

Samir Sadok, Simon Leglaive, Laurent Girin et al.

This article introduces AnCoGen, a novel method that leverages a masked autoencoder to unify the analysis, control, and generation of speech signals within a single model. AnCoGen can analyze speech by estimating key attributes, such as speaker identity, pitch, content, loudness, signal-to-noise ratio, and clarity index. In addition, it can generate speech from these attributes and allow precise control of the synthesized speech by modifying them. Extensive experiments demonstrated the effectiveness of AnCoGen across speech analysis-resynthesis, pitch estimation, pitch modification, and speech enhancement.

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.

Xiaoyu Lin, Laurent Girin, Xavier Alameda-Pineda

In this paper, we propose a latent-variable generative model called mixture of dynamical variational autoencoders (MixDVAE) to model the dynamics of a system composed of multiple moving sources. A DVAE model is pre-trained on a single-source dataset to capture the source dynamics. Then, multiple instances of the pre-trained DVAE model are integrated into a multi-source mixture model with a discrete observation-to-source assignment latent variable. The posterior distributions of both the discrete observation-to-source assignment variable and the continuous DVAE variables representing the sources content/position are estimated using a variational expectation-maximization algorithm, leading to multi-source trajectories estimation. We illustrate the versatility of the proposed MixDVAE model on two tasks: a computer vision task, namely multi-object tracking, and an audio processing task, namely single-channel audio source separation. Experimental results show that the proposed method works well on these two tasks, and outperforms several baseline methods.

Joel Alberto Santos, Zongwei Wu, Xavier Alameda-Pineda et al.

Understanding human instructions is essential for enabling smooth human-robot interaction. In this work, we focus on object grounding, i.e., localizing an object of interest in a visual scene (e.g., an image) based on verbal human instructions. Despite recent progress, a dominant research trend relies on using text as an intermediate representation. These approaches typically transcribe speech to text, extract relevant object keywords, and perform grounding using models pretrained on large text-vision datasets. However, we question both the efficiency and robustness of such transcription-based pipelines. Specifically, we ask: Can we achieve direct audio-visual alignment without relying on text? To explore this possibility, we simplify the task by focusing on grounding from single-word spoken instructions. We introduce a new audio-based grounding dataset that covers a wide variety of objects and diverse human accents. We then adapt and benchmark several models from the closely audio-visual field. Our results demonstrate that direct grounding from audio is not only feasible but, in some cases, even outperforms transcription-based methods, especially in terms of robustness to linguistic variability. Our findings encourage a renewed interest in direct audio grounding and pave the way for more robust and efficient multimodal understanding systems.

Mostafa Sadeghi, Jean-Eudes Ayilo, Romain Serizel et al.

We explore unsupervised speech enhancement using diffusion models as expressive generative priors for clean speech. Existing approaches guide the reverse diffusion process using noisy speech through an approximate, noise-perturbed likelihood score, combined with the unconditional score via a trade-off hyperparameter. In this work, we propose two alternative algorithms that directly model the conditional reverse transition distribution of diffusion states. The first method integrates the diffusion prior with the observation model in a principled way, removing the need for hyperparameter tuning. The second defines a diffusion process over the noisy speech itself, yielding a fully tractable and exact likelihood score. Experiments on the WSJ0-QUT and VoiceBank-DEMAND datasets demonstrate improved enhancement metrics and greater robustness to domain shifts compared to both supervised and unsupervised baselines.

Samir Sadok, Simon Leglaive, Laurent Girin et al.

In this paper, we present a multimodal and dynamical VAE (MDVAE) applied to unsupervised audio-visual speech representation learning. The latent space is structured to dissociate the latent dynamical factors that are shared between the modalities from those that are specific to each modality. A static latent variable is also introduced to encode the information that is constant over time within an audiovisual speech sequence. The model is trained in an unsupervised manner on an audiovisual emotional speech dataset, in two stages. In the first stage, a vector quantized VAE (VQ-VAE) is learned independently for each modality, without temporal modeling. The second stage consists in learning the MDVAE model on the intermediate representation of the VQ-VAEs before quantization. The disentanglement between static versus dynamical and modality-specific versus modality-common information occurs during this second training stage. Extensive experiments are conducted to investigate how audiovisual speech latent factors are encoded in the latent space of MDVAE. These experiments include manipulating audiovisual speech, audiovisual facial image denoising, and audiovisual speech emotion recognition. The results show that MDVAE effectively combines the audio and visual information in its latent space. They also show that the learned static representation of audiovisual speech can be used for emotion recognition with few labeled data, and with better accuracy compared with unimodal baselines and a state-of-the-art supervised model based on an audiovisual transformer architecture.

Xiaoyu Lin, Laurent Girin, Xavier Alameda-Pineda

In this paper, we present an unsupervised probabilistic model and associated estimation algorithm for multi-object tracking (MOT) based on a dynamical variational autoencoder (DVAE), called DVAE-UMOT. The DVAE is a latent-variable deep generative model that can be seen as an extension of the variational autoencoder for the modeling of temporal sequences. It is included in DVAE-UMOT to model the objects' dynamics, after being pre-trained on an unlabeled synthetic dataset of single-object trajectories. Then the distributions and parameters of DVAE-UMOT are estimated on each multi-object sequence to track using the principles of variational inference: Definition of an approximate posterior distribution of the latent variables and maximization of the corresponding evidence lower bound of the data likehood function. DVAE-UMOT is shown experimentally to compete well with and even surpass the performance of two state-of-the-art probabilistic MOT models. Code and data are publicly available.

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.

Guanglei Yang, Enrico Fini, Dan Xu et al.

Over the past years, semantic segmentation, as many other tasks in computer vision, benefited from the progress in deep neural networks, resulting in significantly improved performance. However, deep architectures trained with gradient-based techniques suffer from catastrophic forgetting, which is the tendency to forget previously learned knowledge while learning new tasks. Aiming at devising strategies to counteract this effect, incremental learning approaches have gained popularity over the past years. However, the first incremental learning methods for semantic segmentation appeared only recently. While effective, these approaches do not account for a crucial aspect in pixel-level dense prediction problems, i.e. the role of attention mechanisms. To fill this gap, in this paper we introduce a novel attentive feature distillation approach to mitigate catastrophic forgetting while accounting for semantic spatial- and channel-level dependencies. Furthermore, we propose a {continual attentive fusion} structure, which takes advantage of the attention learned from the new and the old tasks while learning features for the new task. Finally, we also introduce a novel strategy to account for the background class in the distillation loss, thus preventing biased predictions. We demonstrate the effectiveness of our approach with an extensive evaluation on Pascal-VOC 2012 and ADE20K, setting a new state of the art.

Enrico Fini, Victor G. Turrisi da Costa, Xavier Alameda-Pineda et al.

Self-supervised models have been shown to produce comparable or better visual representations than their supervised counterparts when trained offline on unlabeled data at scale. However, their efficacy is catastrophically reduced in a Continual Learning (CL) scenario where data is presented to the model sequentially. In this paper, we show that self-supervised loss functions can be seamlessly converted into distillation mechanisms for CL by adding a predictor network that maps the current state of the representations to their past state. This enables us to devise a framework for Continual self-supervised visual representation Learning that (i) significantly improves the quality of the learned representations, (ii) is compatible with several state-of-the-art self-supervised objectives, and (iii) needs little to no hyperparameter tuning. We demonstrate the effectiveness of our approach empirically by training six popular self-supervised models in various CL settings.

David Emukpere, Xavier Alameda-Pineda, Chris Reinke

A longstanding goal in reinforcement learning is to build intelligent agents that show fast learning and a flexible transfer of skills akin to humans and animals. This paper investigates the integration of two frameworks for tackling those goals: episodic control and successor features. Episodic control is a cognitively inspired approach relying on episodic memory, an instance-based memory model of an agent's experiences. Meanwhile, successor features and generalized policy improvement (SF&GPI) is a meta and transfer learning framework allowing to learn policies for tasks that can be efficiently reused for later tasks which have a different reward function. Individually, these two techniques have shown impressive results in vastly improving sample efficiency and the elegant reuse of previously learned policies. Thus, we outline a combination of both approaches in a single reinforcement learning framework and empirically illustrate its benefits.

Chris Reinke, Xavier Alameda-Pineda

Transfer in Reinforcement Learning aims to improve learning performance on target tasks using knowledge from experienced source tasks. Successor Representations (SR) and their extension Successor Features (SF) are prominent transfer mechanisms in domains where reward functions change between tasks. They reevaluate the expected return of previously learned policies in a new target task to transfer their knowledge. The SF framework extended SR by linearly decomposing rewards into successor features and a reward weight vector allowing their application in high-dimensional tasks. But this came with the cost of having a linear relationship between reward functions and successor features, limiting its application to tasks where such a linear relationship exists. We propose a novel formulation of SR based on learning the cumulative discounted probability of successor features, called Successor Feature Representations (SFR). Crucially, SFR allows to reevaluate the expected return of policies for general reward functions. We introduce different SFR variations, prove its convergence, and provide a guarantee on its transfer performance. Experimental evaluations based on SFR with function approximation demonstrate its advantage over SF not only for general reward functions, but also in the case of linearly decomposable reward functions.

Xiaoyu Bie, Simon Leglaive, Xavier Alameda-Pineda et al.

Dynamical variational autoencoders (DVAEs) are a class of deep generative models with latent variables, dedicated to model time series of high-dimensional data. DVAEs can be considered as extensions of the variational autoencoder (VAE) that include temporal dependencies between successive observed and/or latent vectors. Previous work has shown the interest of using DVAEs over the VAE for speech spectrograms modeling. Independently, the VAE has been successfully applied to speech enhancement in noise, in an unsupervised noise-agnostic set-up that requires neither noise samples nor noisy speech samples at training time, but only requires clean speech signals. In this paper, we extend these works to DVAE-based single-channel unsupervised speech enhancement, hence exploiting both speech signals unsupervised representation learning and dynamics modeling. We propose an unsupervised speech enhancement algorithm that combines a DVAE speech prior pre-trained on clean speech signals with a noise model based on nonnegative matrix factorization, and we derive a variational expectation-maximization (VEM) algorithm to perform speech enhancement. The algorithm is presented with the most general DVAE formulation and is then applied with three specific DVAE models to illustrate the versatility of the framework. Experimental results show that the proposed DVAE-based approach outperforms its VAE-based counterpart, as well as several supervised and unsupervised noise-dependent baselines, especially when the noise type is unseen during training.

Xiaoyu Bie, Laurent Girin, Simon Leglaive et al.

The Variational Autoencoder (VAE) is a powerful deep generative model that is now extensively used to represent high-dimensional complex data via a low-dimensional latent space learned in an unsupervised manner. In the original VAE model, input data vectors are processed independently. In recent years, a series of papers have presented different extensions of the VAE to process sequential data, that not only model the latent space, but also model the temporal dependencies within a sequence of data vectors and corresponding latent vectors, relying on recurrent neural networks. We recently performed a comprehensive review of those models and unified them into a general class called Dynamical Variational Autoencoders (DVAEs). In the present paper, we present the results of an experimental benchmark comparing six of those DVAE models on the speech analysis-resynthesis task, as an illustration of the high potential of DVAEs for speech modeling.

Wen Guo, Xiaoyu Bie, Xavier Alameda-Pineda et al.

Human motion prediction aims to forecast future poses given a sequence of past 3D skeletons. While this problem has recently received increasing attention, it has mostly been tackled for single humans in isolation. In this paper, we explore this problem when dealing with humans performing collaborative tasks, we seek to predict the future motion of two interacted persons given two sequences of their past skeletons. We propose a novel cross interaction attention mechanism that exploits historical information of both persons, and learns to predict cross dependencies between the two pose sequences. Since no dataset to train such interactive situations is available, we collected ExPI (Extreme Pose Interaction), a new lab-based person interaction dataset of professional dancers performing Lindy-hop dancing actions, which contains 115 sequences with 30K frames annotated with 3D body poses and shapes. We thoroughly evaluate our cross interaction network on ExPI and show that both in short- and long-term predictions, it consistently outperforms state-of-the-art methods for single-person motion prediction.

Louis Airale, Dominique Vaufreydaz, Xavier Alameda-Pineda

Prediction of human actions in social interactions has important applications in the design of social robots or artificial avatars. In this paper, we focus on a unimodal representation of interactions and propose to tackle interaction generation in a data-driven fashion. In particular, we model human interaction generation as a discrete multi-sequence generation problem and present SocialInteractionGAN, a novel adversarial architecture for conditional interaction generation. Our model builds on a recurrent encoder-decoder generator network and a dual-stream discriminator, that jointly evaluates the realism of interactions and individual action sequences and operates at different time scales. Crucially, contextual information on interacting participants is shared among agents and reinjected in both the generation and the discriminator evaluation processes. Experiments show that albeit dealing with low dimensional data, SocialInteractionGAN succeeds in producing high realism action sequences of interacting people, comparing favorably to a diversity of recurrent and convolutional discriminator baselines, and we argue that this work will constitute a first stone towards higher dimensional and multimodal interaction generation. Evaluations are conducted using classical GAN metrics, that we specifically adapt for discrete sequential data. Our model is shown to properly learn the dynamics of interaction sequences, while exploiting the full range of available actions.

Mostafa Sadeghi, Xavier Alameda-Pineda

Recently, audio-visual speech enhancement has been tackled in the unsupervised settings based on variational auto-encoders (VAEs), where during training only clean data is used to train a generative model for speech, which at test time is combined with a noise model, e.g. nonnegative matrix factorization (NMF), whose parameters are learned without supervision. Consequently, the proposed model is agnostic to the noise type. When visual data are clean, audio-visual VAE-based architectures usually outperform the audio-only counterpart. The opposite happens when the visual data are corrupted by clutter, e.g. the speaker not facing the camera. In this paper, we propose to find the optimal combination of these two architectures through time. More precisely, we introduce the use of a latent sequential variable with Markovian dependencies to switch between different VAE architectures through time in an unsupervised manner: leading to switching variational auto-encoder (SwVAE). We propose a variational factorization to approximate the computationally intractable posterior distribution. We also derive the corresponding variational expectation-maximization algorithm to estimate the parameters of the model and enhance the speech signal. Our experiments demonstrate the promising performance of SwVAE.

Dan Xu, Xavier Alameda-Pineda, Wanli Ouyang et al.

Multi-scale representations deeply learned via convolutional neural networks have shown tremendous importance for various pixel-level prediction problems. In this paper we present a novel approach that advances the state of the art on pixel-level prediction in a fundamental aspect, i.e. structured multi-scale features learning and fusion. In contrast to previous works directly considering multi-scale feature maps obtained from the inner layers of a primary CNN architecture, and simply fusing the features with weighted averaging or concatenation, we propose a probabilistic graph attention network structure based on a novel Attention-Gated Conditional Random Fields (AG-CRFs) model for learning and fusing multi-scale representations in a principled manner. In order to further improve the learning capacity of the network structure, we propose to exploit feature dependant conditional kernels within the deep probabilistic framework. Extensive experiments are conducted on four publicly available datasets (i.e. BSDS500, NYUD-V2, KITTI, and Pascal-Context) and on three challenging pixel-wise prediction problems involving both discrete and continuous labels (i.e. monocular depth estimation, object contour prediction, and semantic segmentation). Quantitative and qualitative results demonstrate the effectiveness of the proposed latent AG-CRF model and the overall probabilistic graph attention network with feature conditional kernels for structured feature learning and pixel-wise prediction.

Wen Guo, Enric Corona, Francesc Moreno-Noguer et al.

Recent literature addressed the monocular 3D pose estimation task very satisfactorily. In these studies, different persons are usually treated as independent pose instances to estimate. However, in many every-day situations, people are interacting, and the pose of an individual depends on the pose of his/her interactees. In this paper, we investigate how to exploit this dependency to enhance current - and possibly future - deep networks for 3D monocular pose estimation. Our pose interacting network, or PI-Net, inputs the initial pose estimates of a variable number of interactees into a recurrent architecture used to refine the pose of the person-of-interest. Evaluating such a method is challenging due to the limited availability of public annotated multi-person 3D human pose datasets. We demonstrate the effectiveness of our method in the MuPoTS dataset, setting the new state-of-the-art on it. Qualitative results on other multi-person datasets (for which 3D pose ground-truth is not available) showcase the proposed PI-Net. PI-Net is implemented in PyTorch and the code will be made available upon acceptance of the paper.

Laurent Girin, Simon Leglaive, Xiaoyu Bie et al.

Variational autoencoders (VAEs) are powerful deep generative models widely used to represent high-dimensional complex data through a low-dimensional latent space learned in an unsupervised manner. In the original VAE model, the input data vectors are processed independently. Recently, a series of papers have presented different extensions of the VAE to process sequential data, which model not only the latent space but also the temporal dependencies within a sequence of data vectors and corresponding latent vectors, relying on recurrent neural networks or state-space models. In this paper, we perform a literature review of these models. We introduce and discuss a general class of models, called dynamical variational autoencoders (DVAEs), which encompasses a large subset of these temporal VAE extensions. Then, we present in detail seven recently proposed DVAE models, with an aim to homogenize the notations and presentation lines, as well as to relate these models with existing classical temporal models. We have reimplemented those seven DVAE models and present the results of an experimental benchmark conducted on the speech analysis-resynthesis task (the PyTorch code is made publicly available). The paper concludes with a discussion on important issues concerning the DVAE class of models and future research guidelines.

Viet-Nhat Nguyen, Mostafa Sadeghi, Elisa Ricci et al.

In this paper, we are interested in audio-visual speech separation given a single-channel audio recording as well as visual information (lips movements) associated with each speaker. We propose an unsupervised technique based on audio-visual generative modeling of clean speech. More specifically, during training, a latent variable generative model is learned from clean speech spectrograms using a variational auto-encoder (VAE). To better utilize the visual information, the posteriors of the latent variables are inferred from mixed speech (instead of clean speech) as well as the visual data. The visual modality also serves as a prior for latent variables, through a visual network. At test time, the learned generative model (both for speaker-independent and speaker-dependent scenarios) is combined with an unsupervised non-negative matrix factorization (NMF) variance model for background noise. All the latent variables and noise parameters are then estimated by a Monte Carlo expectation-maximization algorithm. Our experiments show that the proposed unsupervised VAE-based method yields better separation performance than NMF-based approaches as well as a supervised deep learning-based technique.

Yahui Liu, Marco De Nadai, Deng Cai et al.

Manipulating visual attributes of images through human-written text is a very challenging task. On the one hand, models have to learn the manipulation without the ground truth of the desired output. On the other hand, models have to deal with the inherent ambiguity of natural language. Previous research usually requires either the user to describe all the characteristics of the desired image or to use richly-annotated image captioning datasets. In this work, we propose a novel unsupervised approach, based on image-to-image translation, that alters the attributes of a given image through a command-like sentence such as "change the hair color to black". Contrarily to state-of-the-art approaches, our model does not require a human-annotated dataset nor a textual description of all the attributes of the desired image, but only those that have to be modified. Our proposed model disentangles the image content from the visual attributes, and it learns to modify the latter using the textual description, before generating a new image from the content and the modified attribute representation. Because text might be inherently ambiguous (blond hair may refer to different shadows of blond, e.g. golden, icy, sandy), our method generates multiple stochastic versions of the same translation. Experiments show that the proposed model achieves promising performances on two large-scale public datasets: CelebA and CUB. We believe our approach will pave the way to new avenues of research combining textual and speech commands with visual attributes.

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.

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/.

Yahui Liu, Marco De Nadai, Jian Yao et al.

Unsupervised image-to-image translation (UNIT) aims at learning a mapping between several visual domains by using unpaired training images. Recent studies have shown remarkable success for multiple domains but they suffer from two main limitations: they are either built from several two-domain mappings that are required to be learned independently, or they generate low-diversity results, a problem known as mode collapse. To overcome these limitations, we propose a method named GMM-UNIT, which is based on a content-attribute disentangled representation where the attribute space is fitted with a GMM. Each GMM component represents a domain, and this simple assumption has two prominent advantages. First, it can be easily extended to most multi-domain and multi-modal image-to-image translation tasks. Second, the continuous domain encoding allows for interpolation between domains and for extrapolation to unseen domains and translations. Additionally, we show how GMM-UNIT can be constrained down to different methods in the literature, meaning that GMM-UNIT is a unifying framework for unsupervised image-to-image translation.