Recent learning-based face reconstruction and registration frameworks such as ToFu and TEMPEH have shown that dense correspondence between facial scans and a common topology can be learned directly from images. However, these approaches still depend on precomputed registrations obtained through iterative optimization pipelines that often require manual verification and correction by human annotators. We introduce MOCHI (Multi-view Optimizable Correspondence of Heads from Images), a fully differentiable and registration-free alternative. Instead of relying on optimization-based registrations, we employ a pseudo-linear inverse kinematic solver in conjunction with dense 2D keypoints produced by a tracker trained only on synthetic data to directly enforce a common face topology at the vertex level. We further find that the commonly used point-to-surface distance can lead to unstable training and artifacts, and instead use pointmap- and normal-based losses that provide smoother gradients, more stable optimization, and improved reconstruction results.Additionally, we introduce at inference a brief test-time-optimization scheme which can further refine the results of the network, resulting in registrations that outperform traditional labor-intensive pipelines.Despite removing external registrations, our extensive experimental results show that MOCHI surpasses the previous state-of-the-art in reconstruction accuracy and visual fidelity. The code and the model will be made public.

Accurate 6D pose estimation for robotic harvesting is fundamentally hindered by the biological deformability and high intra-class shape variability of agricultural produce. Instance-level methods fail in this setting, as obtaining exact 3D models for every unique piece of produce is practically infeasible, while category-level approaches that rely on a fixed template suffer significant accuracy degradation when the prior deviates from the true instance geometry. To bridge such lack of robustness to deformation, we introduce PEAR (Pose and dEformation of Agricultural pRoduce), the first benchmark providing joint 6D pose and per-instance 3D deformation ground truth across 8 produce categories, acquired via a robotic manipulator for high annotation accuracy. Using PEAR, we show that state-of-the-art methods suffer up to 6x performance degradation when faced with the inherent geometric deviations of real-world produce. Motivated by this finding, we propose SEED (Simultaneous Estimation of posE and Deformation), a unified RGB-only framework that jointly predicts 6D pose and explicit lattice deformations from a single image across multiple produce categories. Trained entirely on synthetic data with generative texture augmentation applied at the UV level, SEED outperforms MegaPose on 6 out of 8 categories under identical RGB-only conditions, demonstrating that explicit shape modeling is a critical step toward reliable pose estimation in agricultural robotics.

Monocular head pose estimation is traditionally formulated as direct regression from a single image to an absolute pose. This paradigm forces the network to implicitly internalize a dataset-specific canonical reference frame. In this work, we argue that predicting the relative rigid transformation between two observed head configurations is a fundamentally easier and more robust formulation. We introduce VGGT-HPE, a relative head pose estimator built upon a general-purpose geometry foundation model. Finetuned exclusively on synthetic facial renderings, our method sidesteps the need for an implicit anchor by reducing the problem to estimating a geometric displacement from an explicitly provided anchor with a known pose. As a practical benefit, the relative formulation also allows the anchor to be chosen at test time - for instance, a near-neutral frame or a temporally adjacent one - so that the prediction difficulty can be controlled by the application. Despite zero real-world training data, VGGT-HPE achieves state-of-the-art results on the BIWI benchmark, outperforming established absolute regression methods trained on mixed and real datasets. Through controlled easy- and hard-pair benchmarks, we also systematically validate our core hypothesis: relative prediction is intrinsically more accurate than absolute regression, with the advantage scaling alongside the difficulty of the target pose. Project page and code: https://vasilikivas.github.io/VGGT-HPE

We propose a contrastive self-supervised learning framework to improve transfer learning across dissimilar bioacoustic datasets. Amplitude - and frequency - modulated (AM-FM) delphinid vocalizations from the Watkins Marine Mammal Sound Database (WMMSD) are used for pretraining, while two bird-call datasets -Rain Forest Connection Species Audio Detection (RFCx - bird and frog calls) and BirdCLEF 2022 (bird vocalizations)-serve for downstream evaluation. We train ResNet18, MobileNetV2, and Vision Transformer (ViT-B/16) backbones, in a two-stage pipeline. In the pretraining stage, the model contrasts different time-frequency representations of the same audio, such as spectrogram-energogram pairs derived via the discrete or regularized Teager-Kaiser Energy Operator (TKEO). In the fine-tuning stage, the pretrained models are adapted for bird vocalization classification. Experiments show that representation-diverse contrastive learning on delphinid calls, followed by fine-tuning, consistently outperforms both fully supervised baselines and conventional self-supervised approaches based solely on augmented spectrograms. The best results are obtained when contrasting spectrograms with TKEO-based energograms, indicating that cross-representation contrastive learning yields more robust and transferable embeddings for bioacoustic tasks.

Existing Gaussian-splatting-based monocular Simultaneous Localization and Mapping (SLAM) systems are either tailored to short sequences, are not real-time, or suffer from prohibitive GPU memory requirements, limiting their applicability in realistic, long-horizon scenarios. To address this, we present GLAM-SLAM, a real-time, decoupled Gaussian-splatting SLAM system designed for large-scale outdoor scenes. We ensure lightweight tracking using a robust, feature-based SLAM frontend, while for mapping, we adopt a structured, sparse anchor grid representation that ensures scalable operation and maintains scene coherence across long-term sequences. To satisfy the dense initialization requirements of 3D Gaussian Splatting (3DGS), we introduce a geometry-based flow-densification anchoring strategy using epipolar constraints. Furthermore, by treating mapping as a multi-scene problem, we propose a scene-partitioning strategy that introduces a strong spatial inductive bias via MLP initializations to generate localized Gaussians. We evaluate our system on the challenging, long-sequence KITTI Odometry, Oxford RobotCar, and M'alaga datasets. Extensive ablations and comparisons demonstrate a 15

Recent frameworks like ToFu and TEMPEH provide an automated alternative to classical registration pipelines by predicting 3D meshes in dense semantic correspondence directly from calibrated multi-view images. However, these learning-based methods rely on the slow, manual registration pipelines they aim to replace for their training supervision. We overcome this limitation with MOCHI (Multi-view Optimizable Correspondence of Heads from Images), a multi-view 3D face prediction framework trained without requiring registered training data. MOCHI eliminates the registration data dependency by enforcing topological consistency through a pseudo-linear inverse kinematic solver. Semantic alignment is guided by dense keypoints from a 2D landmark predictor trained exclusively on synthetic data. Our analysis further reveals that standard point-to-surface distances induce training instabilities and visual artifacts in registration-free settings. We propose pointmap- and normal-based losses instead, which provide smoother gradients and superior reconstruction fidelity. Finally, we introduce a test-time optimization scheme that refines network weights over a few dozen iterations. This approach bridges the gap between feed-forward efficiency and iterative optimization precision, allowing MOCHI to outperform traditional labor-intensive pipelines in both reconstruction accuracy and visual quality. Code and model are public at: https://filby89.github.io/mochi.

Expanding neural networks during training is a promising way to augment capacity without retraining larger models from scratch. However, newly added neurons often fail to adjust to a trained network and become inactive, providing no contribution to capacity growth. We propose the Shared-Weights Extender (SWE), a novel method explicitly designed to prevent inactivity of new neurons by coupling them with existing ones for smooth integration. In parallel, we introduce the Steepest Voting Distributor (SVoD), a gradient-based method for allocating neurons across layers during deep network expansion. Our extensive benchmarking on four datasets shows that our method can effectively suppress neuron inactivity and achieve better performance compared to other expanding methods and baselines.

Digital phenotyping enables continuous passive monitoring of behavior and physiology, offering a promising paradigm for early detection of psychotic relapse. In this work, we develop and systematically study two smartwatch-based frameworks for daily relapse detection. The first forecasts cardiac dynamics and flags deviations between predicted and observed features as indicators of abnormality. The second adopts a multi-task formulation that fuses sleep with motion and cardiac-derived signals, learning time-aware embeddings and predicting measurement timing. Both pipelines use Transformer encoders and output a daily anomaly score, derived from predictive uncertainty estimated via an ensemble of multilayer perceptrons to improve robustness to real-world wearable variability. While each framework independently demonstrates strong predictive power, we show that they capture complementary physiological signatures. Consequently, we propose a late-fusion strategy that synergistically combines the anomaly signals from both architectures into a unified decision score. We benchmark our methodology on the 2nd e-Prevention Grand Challenge dataset, where our fused model achieves a 8

The availability of large, unlabeled datasets across various domains has contributed to the development of a plethora of methods that learn representations for multiple target (downstream) tasks through self-supervised pre-training. In this work, we introduce CVSM (Contrastive Vocal Similarity Modeling), a contrastive self-supervised procedure for music signal representation learning in the audio domain that can be utilized for musical and vocal similarity modeling. Our method operates under a contrastive framework, maximizing the similarity between vocal excerpts and musical mixtures containing the same vocals; we devise both a label-informed protocol, leveraging artist identity information to sample the contrastive pairs, and a label-agnostic scheme, involving artificial mixture creation from randomly sampled vocal and accompaniment excerpts, which are paired with vocals from the same audio segment. We evaluate our proposed method in measuring vocal similarity both objectively, through linear probing on a suite of appropriate downstream tasks, and subjectively, via conducting a user study consisting of pairwise comparisons between different models in a recommendation-by-query setting. Our results indicate that the representations learned through CVSM are effective in musical and vocal similarity modeling, outperforming numerous baselines across both isolated vocals and complete musical mixtures. Moreover, while the availability of artist identity labels during pre-training leads to overall more consistent performance both in the evaluated downstream tasks and the user study, a label-agnostic CVSM variant incorporating hybrid pre-training with real and artificial mixtures achieves comparable performance to the label-informed one in artist identification and perceived vocal similarity.

Inspection of ultra-high voltage substations (UHVS) plays a crucial role for ensuring the stability of power grids and preventing damage that can potentially lead to serious power loss. Automating the inspection process provides early detection of equipment faults, improved safety and cost efficiency. This paper presents an automated visual inspection system, using an RGB and thermal camera mounted on an autonomous ground robot. Our approach leverages component detection in RGB images with YOLOvl ln, followed by multi-modal image matching to locate components in thermal imagery and a rule-based anomaly detection algorithm. Experimental results in an operational power substation demonstrate the system's ability to detect early-stage thermal anomalies, highlighting its potential for improving substation reliability and operational safety. Copyright (c) 2025 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

Visibility maps are crucial for autonomous robotic applications such as exploration, path planning, obstacle avoidance, and multi-robot coordination. In the context of electrical transmission infrastructure, automated robotic inspection enhances proactive maintenance, enabling early detection of wear, damage, or faults, thereby improving safety, extending component lifespan, and optimizing maintenance schedules.In this work, we propose an algorithm to compute optimal visibility locations, enabling a mobile robot to acquire RGB and thermal images for fault detection. Using LiDAR scans, we construct a global 3D map composed of ground structures (represented as a Grid Map) and overground structures (modeled with an OctoMap for efficient ray-casting). We apply clustering techniques to identify 3D bounding boxes for electrical components and define suitable source and target points for visibility assessment. By employing a weighted visibility scoring approach, we determine the ground positions that offer the best visibility of each component while ensuring minimal occlusions and adherence to viewing constraints.The proposed method enables a robot to autonomously navigate to these optimal viewpoints, improving inspection efficiency. By integrating visibility regions across multiple components, the inspection process is further optimized, reducing overall inspection time. Our algorithm has been successfully deployed and tested at an Ultra-High Voltage Center (UHVC) in Greece, demonstrating its effectiveness in real-world conditions.

Navigating indoor environments poses significant challenges for individuals with mobility and cognitive impairments, impacting their independence and quality of life. While robotic rollators have shown potential in providing mobility assistance, existing systems often rely on static navigation approaches that lack a capacity to dynamically adapt to path changes and unforeseen deviations. This paper addresses this gap by introducing a dynamic turn-by-turn navigation system designed to provide real-time, user-centered guidance. The system aims to enhance spatial orientation and wayfinding through online dynamic planning, addressing critical limitations of current solutions. We present results from user trials in a simulated environment, which show its potential to improve mobility. Possible areas of improvement are also discussed, paving the way for more effective assistive technologies in real-world settings.

The progress of composed image retrieval (CIR), a popular research direction in image retrieval, where a combined visual and textual query is used, is held back by the absence of high-quality training and evaluation data. We introduce a new evaluation dataset, i-CIR, which, unlike existing datasets, focuses on an instance-level class definition. The goal is to retrieve images that contain the same particular object as the visual query, presented under a variety of modifications defined by textual queries. Its design and curation process keep the dataset compact to facilitate future research, while maintaining its challenge—comparable to retrieval among more than 40M random distractors—through a semi-automated selection of hard negatives. To overcome the challenge of obtaining clean, diverse, and suitable training data, we leverage pre-trained vision-and-language models (VLMs) in a training-free approach called BASIC. The method separately estimates query-image-to-image and query-text-to-image similarities, performing late fusion to upweight images that satisfy both queries, while down-weighting those that exhibit high similarity with only one of the two. Each individual similarity is further improved by a set of components that are simple and intuitive. BASIC sets a new state of the art on i-CIR but also on existing CIR datasets that follow a semantic-level class definition. Project page: https://vrg.fel.cvut.cz/icir/.

We propose Diffusion-Informed Model Predictive Control (D-I MPC), a generic framework for uncertainty-aware prediction and decision-making in partially observable stochastic systems by integrating diffusion-based time series forecasting models in Model Predictive Control algorithms. In our approach, a diffusion-based time series forecasting model is used to probabilistically estimate the evolution of the system's stochastic components. These forecasts are then incorporated into MPC algorithms to estimate future trajectories and optimize action selection under the uncertainty of the future. We evaluate the framework on the task of energy arbitrage, where a Battery Energy Storage System participates in the day-ahead electricity market of the New York state. Experimental results indicate that our model-based approach with a diffusion-based forecaster significantly outperforms both implementations with classical forecasting methods and model-free reinforcement learning baselines.

We propose GHR-VQA, Graph-guided Hierarchical Relational Reasoning for Video Question Answering (Video QA), a novel human-centric framework that incorporates scene graphs to capture intricate human-object interactions within video sequences. Unlike traditional pixel-based methods, each frame is represented as a scene graph and human nodes across frames are linked to a global root, forming the video-level graph and enabling cross-frame reasoning centered on human actors. The video-level graphs are then processed by Graph Neural Networks (GNNs), transforming them into rich, context-aware embeddings for efficient processing. Finally, these embeddings are integrated with question features in a hierarchical network operating across different abstraction levels, enhancing both local and global understanding of video content. This explicit human-rooted structure enhances interpretability by decomposing actions into human-object interactions and enables a more profound understanding of spatiotemporal dynamics. We validate our approach on the Action Genome Question Answering (AGQA) dataset, achieving significant performance improvements, including a 7.3

We investigate deep morphological neural networks (DMNNs). We demonstrate that despite their inherent non-linearity, "linear" activations are essential for DMNNs. To preserve their inherent sparsity, we propose architectures that constraint the parameters of the "linear" activations: For the first (resp. second) architecture, we work under the constraint that the majority of parameters (resp. learnable parameters) should be part of morphological operations. We improve the generalization ability of our networks via residual connections and weight dropout. Our proposed networks can be successfully trained, and are more prunable than linear networks. To the best of our knowledge, we are the first to successfully train DMNNs under such constraints. Finally, we propose a hybrid network architecture combining linear and morphological layers, showing empirically that the inclusion of morphological layers significantly accelerates the convergence of gradient descent with large batches.

Inspired by infant development, we propose a Reinforcement Learning (RL) framework for autonomous self-exploration in a robotic agent, Baby Sophia, using the BabyBench simulation environment. The agent learns self-touch and hand regard behaviors through intrinsic rewards that mimic an infant's curiosity-driven exploration of its own body. For self-touch, high-dimensional tactile inputs are transformed into compact, meaningful representations, enabling efficient learning. The agent then discovers new tactile contacts through intrinsic rewards and curriculum learning that encourage broad body coverage, balance, and generalization. For hand regard, visual features of the hands, such as skin-color and shape, are learned through motor babbling. Then, intrinsic rewards encourage the agent to perform novel hand motions, and follow its hands with its gaze. A curriculum learning setup from single-hand to dual-hand training allows the agent to reach complex visual-motor coordination. The results of this work demonstrate that purely curiosity-based signals, with no external supervision, can drive coordinated multimodal learning, imitating an infant's progression from random motor babbling to purposeful behaviors.

Real-world applications focused on interpreting human behavior often require multiple computer vision tasks to be addressed simultaneously. Multitask learning typically achieves this by jointly training a single deep neural network to learn shared representations, providing efficiency and improving generalization. Although action and gesture recognition are closely related tasks, since they focus on body and hand movements, current state-of-the-art methods handle them separately. In this paper, we show that employing a multi-task learning paradigm for action and gesture recognition results in more efficient, robust and generalizable visual representations, by leveraging the synergies between these tasks. Experiments on multiple action and gesture datasets demonstrate that handling actions and gestures in a single architecture can achieve better performance for both tasks in comparison to their single-task learning variants.

Tropical geometry has recently found several applications in the analysis of neural networks with piecewise linear activation functions. This article presents a new look at the problem of tropical polynomial division and its application to the simplification of neural networks. We analyze tropical polynomials with real coefficients, extending earlier ideas and methods developed for polynomials with integer coefficients. We first prove the existence of a unique quotient-remainder pair and characterize the quotient in terms of the convex bi-conjugate of a related function. Interestingly, the quotient of tropical polynomials with integer coefficients does not necessarily have integer coefficients. Furthermore, we develop a relationship of tropical polynomial division with the computation of the convex hull of unions of convex polyhedra and use it to derive an exact algorithm for tropical polynomial division. An approximate algorithm is also presented, based on an alternation between data partition and linear programming. We also develop special techniques to divide composite polynomials, described as sums or maxima of simpler ones. Finally, we provide numerical results to demonstrate the efficiency of the proposed algorithms, using the MNIST handwritten digits, SVHN, CIFAR-10, and CIFAR-100 datasets, along with an application example in learning model predictive control (LMPC).

Guitar string separation constitutes a source separation task in which the individual stems to be retrieved are the signals produced by each distinct guitar string. Certain hardware solutions for string-wise signal manipulation are gaining traction among the guitarist community for applications such as MIDI control and string-level sound effects. However, software techniques for string separation are lagging behind, even though they could form a competitive alternative. This work applies established deep learning architectures commonly used for standard music source separation, to address the task of guitar string separation. It opts for a waveform-to-waveform approach using a multi-channel version of Wave-U-Net. It further suggests that this model can serve as a bridge from separation to transcription by facilitating guitar tablature inference. Tablature transcription enhancement proved feasible using a method that relies on feeding the separated signals to standard tablature inference models, by simply modifying their initial convolution layer to handle multi-source input. Moreover, this work seeks to address the challenges encountered in the more specific string separation scenario where no target signals are readily available for training, meaning that these signals are not provided directly through sophisticated hardware equipment such as polyphonic pickups. Instead, data manipulation and augmentation techniques that produce sample-level synthesized targets are proposed. Thus, two newly created datasets are introduced, namely GS-Aux and ADGP, based on standard guitar datasets: GuitarSet and DadaGP.