Christophe De Vleeschouwer

Vladimir Somers, Christophe De Vleeschouwer, Alexandre Alahi

Occluded person re-identification (ReID) is a person retrieval task which aims at matching occluded person images with holistic ones. For addressing occluded ReID, part-based methods have been shown beneficial as they offer fine-grained information and are well suited to represent partially visible human bodies. However, training a part-based model is a challenging task for two reasons. Firstly, individual body part appearance is not as discriminative as global appearance (two distinct IDs might have the same local appearance), this means standard ReID training objectives using identity labels are not adapted to local feature learning. Secondly, ReID datasets are not provided with human topographical annotations. In this work, we propose BPBreID, a body part-based ReID model for solving the above issues. We first design two modules for predicting body part attention maps and producing body part-based features of the ReID target. We then propose GiLt, a novel training scheme for learning part-based representations that is robust to occlusions and non-discriminative local appearance. Extensive experiments on popular holistic and occluded datasets show the effectiveness of our proposed method, which outperforms state-of-the-art methods by 0.7% mAP and 5.6% rank-1 accuracy on the challenging Occluded-Duke dataset. Our code is available at https://github.com/VlSomers/bpbreid.

Anthony Cioppa, Silvio Giancola, Vladimir Somers et al. · pku

The SoccerNet 2023 challenges were the third annual video understanding challenges organized by the SoccerNet team. For this third edition, the challenges were composed of seven vision-based tasks split into three main themes. The first theme, broadcast video understanding, is composed of three high-level tasks related to describing events occurring in the video broadcasts: (1) action spotting, focusing on retrieving all timestamps related to global actions in soccer, (2) ball action spotting, focusing on retrieving all timestamps related to the soccer ball change of state, and (3) dense video captioning, focusing on describing the broadcast with natural language and anchored timestamps. The second theme, field understanding, relates to the single task of (4) camera calibration, focusing on retrieving the intrinsic and extrinsic camera parameters from images. The third and last theme, player understanding, is composed of three low-level tasks related to extracting information about the players: (5) re-identification, focusing on retrieving the same players across multiple views, (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams, and (7) jersey number recognition, focusing on recognizing the jersey number of players from tracklets. Compared to the previous editions of the SoccerNet challenges, tasks (2-3-7) are novel, including new annotations and data, task (4) was enhanced with more data and annotations, and task (6) now focuses on end-to-end approaches. More information on the tasks, challenges, and leaderboards are available on https://www.soccer-net.org. Baselines and development kits can be found on https://github.com/SoccerNet.

Karim El Khoury, Maxime Zanella, Benoît Gérin et al.

Vision-Language Models for remote sensing have shown promising uses thanks to their extensive pretraining. However, their conventional usage in zero-shot scene classification methods still involves dividing large images into patches and making independent predictions, i.e., inductive inference, thereby limiting their effectiveness by ignoring valuable contextual information. Our approach tackles this issue by utilizing initial predictions based on text prompting and patch affinity relationships from the image encoder to enhance zero-shot capabilities through transductive inference, all without the need for supervision and at a minor computational cost. Experiments on 10 remote sensing datasets with state-of-the-art Vision-Language Models demonstrate significant accuracy improvements over inductive zero-shot classification. Our source code is publicly available on Github: https://github.com/elkhouryk/RS-TransCLIP

Vladimir Somers, Christophe De Vleeschouwer, Alexandre Alahi

Occluded Person Re-Identification (ReID) is a metric learning task that involves matching occluded individuals based on their appearance. While many studies have tackled occlusions caused by objects, multi-person occlusions remain less explored. In this work, we identify and address a critical challenge overlooked by previous occluded ReID methods: the Multi-Person Ambiguity (MPA) arising when multiple individuals are visible in the same bounding box, making it impossible to determine the intended ReID target among the candidates. Inspired by recent work on prompting in vision, we introduce Keypoint Promptable ReID (KPR), a novel formulation of the ReID problem that explicitly complements the input bounding box with a set of semantic keypoints indicating the intended target. Since promptable re-identification is an unexplored paradigm, existing ReID datasets lack the pixel-level annotations necessary for prompting. To bridge this gap and foster further research on this topic, we introduce Occluded-PoseTrack ReID, a novel ReID dataset with keypoints labels, that features strong inter-person occlusions. Furthermore, we release custom keypoint labels for four popular ReID benchmarks. Experiments on person retrieval, but also on pose tracking, demonstrate that our method systematically surpasses previous state-of-the-art approaches on various occluded scenarios. Our code, dataset and annotations are available at https://github.com/VlSomers/keypoint_promptable_reidentification.

Antoine Vanderschueren, Christophe De Vleeschouwer

Turning the weights to zero when training a neural network helps in reducing the computational complexity at inference. To progressively increase the sparsity ratio in the network without causing sharp weight discontinuities during training, our work combines soft-thresholding and straight-through gradient estimation to update the raw, i.e. non-thresholded, version of zeroed weights. Our method, named ST-3 for straight-through/soft-thresholding/sparse-training, obtains SoA results, both in terms of accuracy/sparsity and accuracy/FLOPS trade-offs, when progressively increasing the sparsity ratio in a single training cycle. In particular, despite its simplicity, ST-3 favorably compares to the most recent methods, adopting differentiable formulations or bio-inspired neuroregeneration principles. This suggests that the key ingredients for effective sparsification primarily lie in the ability to give the weights the freedom to evolve smoothly across the zero state while progressively increasing the sparsity ratio. Source code and weights available at https://github.com/vanderschuea/stthree

Seyed Abolfazl Ghasemzadeh, Alexandre Alahi, Christophe De Vleeschouwer

Estimating 3D human poses from 2D images is challenging due to occlusions and projective acquisition. Learning-based approaches have been largely studied to address this challenge, both in single and multi-view setups. These solutions however fail to generalize to real-world cases due to the lack of (multi-view) 'in-the-wild' images paired with 3D poses for training. For this reason, we propose combining 2D pose estimation, for which large and rich training datasets exist, and 2D-to-3D pose lifting, using a transformer-based network that can be trained from synthetic 2D-3D pose pairs. Our experiments demonstrate decreases up to 45% in MPJPE errors compared to the 3D pose obtained by triangulating the 2D poses. The framework's source code is available at https://github.com/aghasemzadeh/OpenMPL .

Silvio Giancola, Anthony Cioppa, Adrien Deliège et al.

The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.

Benjamin Kiefer, Matej Kristan, Janez Perš et al.

The 1$^{\text{st}}$ Workshop on Maritime Computer Vision (MaCVi) 2023 focused on maritime computer vision for Unmanned Aerial Vehicles (UAV) and Unmanned Surface Vehicle (USV), and organized several subchallenges in this domain: (i) UAV-based Maritime Object Detection, (ii) UAV-based Maritime Object Tracking, (iii) USV-based Maritime Obstacle Segmentation and (iv) USV-based Maritime Obstacle Detection. The subchallenges were based on the SeaDronesSee and MODS benchmarks. This report summarizes the main findings of the individual subchallenges and introduces a new benchmark, called SeaDronesSee Object Detection v2, which extends the previous benchmark by including more classes and footage. We provide statistical and qualitative analyses, and assess trends in the best-performing methodologies of over 130 submissions. The methods are summarized in the appendix. The datasets, evaluation code and the leaderboard are publicly available at https://seadronessee.cs.uni-tuebingen.de/macvi.

Seyed Abolfazl Ghasemzadeh, Alexandre Alahi, Christophe De Vleeschouwer

Estimating 3D human poses from 2D images remains challenging due to occlusions and projective ambiguity. Multi-view learning-based approaches mitigate these issues but often fail to generalize to real-world scenarios, as large-scale multi-view datasets with 3D ground truth are scarce and captured under constrained conditions. To overcome this limitation, recent methods rely on 2D pose estimation combined with 2D-to-3D pose lifting trained on synthetic data. Building on our previous MPL framework, we propose RUMPL, a transformer-based 3D pose lifter that introduces a 3D ray-based representation of 2D keypoints. This formulation makes the model independent of camera calibration and the number of views, enabling universal deployment across arbitrary multi-view configurations without retraining or fine-tuning. A new View Fusion Transformer leverages learned fused-ray tokens to aggregate information along rays, further improving multi-view consistency. Extensive experiments demonstrate that RUMPL reduces MPJPE by up to 53% compared to triangulation and over 60% compared to transformer-based image-representation baselines. Results on new benchmarks, including in-the-wild multi-view and multi-person datasets, confirm its robustness and scalability. The framework's source code is available at https://github.com/aghasemzadeh/OpenRUMPL

Laurent Jacques, Christophe De Vleeschouwer

This paper studies the effect of discretizing the parametrization of a dictionary used for Matching Pursuit decompositions of signals. Our approach relies on viewing the continuously parametrized dictionary as an embedded manifold in the signal space on which the tools of differential (Riemannian) geometry can be applied. The main contribution of this paper is twofold. First, we prove that if a discrete dictionary reaches a minimal density criterion, then the corresponding discrete MP (dMP) is equivalent in terms of convergence to a weakened hypothetical continuous MP. Interestingly, the corresponding weakness factor depends on a density measure of the discrete dictionary. Second, we show that the insertion of a simple geometric gradient ascent optimization on the atom dMP selection maintains the previous comparison but with a weakness factor at least two times closer to unity than without optimization. Finally, we present numerical experiments confirming our theoretical predictions for decomposition of signals and images on regular discretizations of dictionary parametrizations.

Alexandre Englebert, Olivier Cornu, Christophe De Vleeschouwer

The need for Explainable AI is increasing with the development of deep learning. The saliency maps derived from convolutional neural networks generally fail in localizing with accuracy the image features justifying the network prediction. This is because those maps are either low-resolution as for CAM [Zhou et al., 2016], or smooth as for perturbation-based methods [Zeiler and Fergus, 2014], or do correspond to a large number of widespread peaky spots as for gradient-based approaches [Sundararajan et al., 2017, Smilkov et al., 2017]. In contrast, our work proposes to combine the information from earlier network layers with the one from later layers to produce a high resolution Class Activation Map that is competitive with the previous art in term of insertion-deletion faithfulness metrics, while outperforming it in term of precision of class-specific features localization.

Anthony Cioppa, Silvio Giancola, Vladimir Somers et al.

The SoccerNet 2024 challenges represent the fourth annual video understanding challenges organized by the SoccerNet team. These challenges aim to advance research across multiple themes in football, including broadcast video understanding, field understanding, and player understanding. This year, the challenges encompass four vision-based tasks. (1) Ball Action Spotting, focusing on precisely localizing when and which soccer actions related to the ball occur, (2) Dense Video Captioning, focusing on describing the broadcast with natural language and anchored timestamps, (3) Multi-View Foul Recognition, a novel task focusing on analyzing multiple viewpoints of a potential foul incident to classify whether a foul occurred and assess its severity, (4) Game State Reconstruction, another novel task focusing on reconstructing the game state from broadcast videos onto a 2D top-view map of the field. Detailed information about the tasks, challenges, and leaderboards can be found at https://www.soccer-net.org, with baselines and development kits available at https://github.com/SoccerNet.

Marcello Davide Caio, Gabriel Van Zandycke, Christophe De Vleeschouwer

Accurately localizing objects in three dimensions (3D) is crucial for various computer vision applications, such as robotics, autonomous driving, and augmented reality. This task finds another important application in sports analytics and, in this work, we present a novel method for 3D basketball localization from a single calibrated image. Our approach predicts the object's height in pixels in image space by estimating its projection onto the ground plane within the image, leveraging the image itself and the object's location as inputs. The 3D coordinates of the ball are then reconstructed by exploiting the known projection matrix. Extensive experiments on the public DeepSport dataset, which provides ground truth annotations for 3D ball location alongside camera calibration information for each image, demonstrate the effectiveness of our method, offering substantial accuracy improvements compared to recent work. Our work opens up new possibilities for enhanced ball tracking and understanding, advancing computer vision in diverse domains. The source code of this work is made publicly available at \url{https://github.com/gabriel-vanzandycke/deepsport}.

Benoit Brummer, Christophe De Vleeschouwer

Image noise is ubiquitous in photography. However, image noise is not compressible nor desirable, thus attempting to convey the noise in compressed image bitstreams yields sub-par results in both rate and distortion. We propose to explicitly learn the image denoising task when training a codec. Therefore, we leverage the Natural Image Noise Dataset, which offers a wide variety of scenes captured with various ISO numbers, leading to different noise levels, including insignificant ones. Given this training set, we supervise the codec with noisy-clean image pairs, and show that a single model trained based on a mixture of images with variable noise levels appears to yield best-in-class results with both noisy and clean images, achieving better rate-distortion than a compression-only model or even than a pair of denoising-then-compression models with almost one order of magnitude fewer GMac operations.

Antoine Legrand, Benoît Macq, Christophe De Vleeschouwer

JPEG-XS offers low complexity image compression for applications with constrained but reasonable bit-rate, and low latency. Our paper explores the deployment of JPEG-XS on lossy packet networks. To preserve low latency, Forward Error Correction (FEC) is envisioned as the protection mechanism of interest. Despite the JPEG-XS codestream is not scalable in essence, we observe that the loss of a codestream fraction impacts the decoded image quality differently, depending on whether this codestream fraction corresponds to codestream headers, to coefficients significance information, or to low/high frequency data, respectively. Hence, we propose a rate-distortion optimal unequal error protection scheme that adapts the redundancy level of Reed-Solomon codes according to the rate of channel losses and the type of information protected by the code. Our experiments demonstrate that, at 5% loss rates, it reduces the Mean Squared Error by up to 92% and 65%, compared to a transmission without and with optimal but equal protection, respectively.

Niels Sayez, Christophe De Vleeschouwer

Collecting image annotations remains a significant burden when deploying CNN in a specific applicative context. This is especially the case when the annotation consists in binary masks covering object instances. Our work proposes to delineate instances in three steps, based on a semi-automatic approach: (1) the extreme points of an object (left-most, right-most, top, bottom pixels) are manually defined, thereby providing the object bounding-box, (2) a universal automatic segmentation tool like Deep Extreme Cut is used to turn the bounded object into a segmentation mask that matches the extreme points; and (3) the predicted mask is manually corrected. Various strategies are then investigated to balance the human manual annotation resources between bounding-box definition and mask correction, including when the correction of instance masks is prioritized based on their overlap with other instance bounding-boxes, or the outcome of an instance segmentation model trained on a partially annotated dataset. Our experimental study considers a teamsport player segmentation task, and measures how the accuracy of the Panoptic-Deeplab instance segmentation model depends on the human annotation resources allocation strategy. It reveals that the sole definition of extreme points results in a model accuracy that would require up to 10 times more resources if the masks were defined through fully manual delineation of instances. When targeting higher accuracies, prioritizing the mask correction among the training set instances is also shown to save up to 80\% of correction annotation resources compared to a systematic frame by frame correction of instances, for a same trained instance segmentation model accuracy.

Mathieu Cyrille Simon, Pascal Frossard, Christophe De Vleeschouwer

This paper explores self-supervised disentangled representation learning within sequential data, focusing on separating time-independent and time-varying factors in videos. We propose a new model that breaks the usual independence assumption between those factors by explicitly accounting for the causal relationship between the static/dynamic variables and that improves the model expressivity through additional Normalizing Flows. A formal definition of the factors is proposed. This formalism leads to the derivation of sufficient conditions for the ground truth factors to be identifiable, and to the introduction of a novel theoretically grounded disentanglement constraint that can be directly and efficiently incorporated into our new framework. The experiments show that the proposed approach outperforms previous complex state-of-the-art techniques in scenarios where the dynamics of a scene are influenced by its content.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

This work introduces a novel augmentation method that increases the diversity of a train set to improve the generalization abilities of a 6D pose estimation network. For this purpose, a Neural Radiance Field is trained from synthetic images and exploited to generate an augmented set. Our method enriches the initial set by enabling the synthesis of images with (i) unseen viewpoints, (ii) rich illumination conditions through appearance extrapolation, and (iii) randomized textures. We validate our augmentation method on the challenging use-case of spacecraft pose estimation and show that it significantly improves the pose estimation generalization capabilities. On the SPEED+ dataset, our method reduces the error on the pose by 50% on both target domains.

Karim El Khoury, Maxime Zanella, Tiffanie Godelaine et al.

Audio-language models have recently demonstrated strong zero-shot capabilities by leveraging natural-language supervision to classify audio events without labeled training data. Yet, their performance is highly sensitive to the wording of text prompts, with small variations leading to large fluctuations in accuracy. Prior work has mitigated this issue through prompt learning or prompt ensembling. However, these strategies either require annotated data or fail to account for the fact that some prompts may negatively impact performance. In this work, we present an entropy-guided prompt weighting approach that aims to find a robust combination of prompt contributions to maximize prediction confidence. To this end, we formulate a tailored objective function that minimizes prediction entropy to yield new prompt weights, utilizing low-entropy as a proxy for high confidence. Our approach can be applied to individual samples or a batch of audio samples, requiring no additional labels and incurring negligible computational overhead. Experiments on five audio classification datasets covering environmental, urban, and vocal sounds, demonstrate consistent gains compared to classical prompt ensembling methods in a zero-shot setting, with accuracy improvements 5-times larger across the whole benchmark.

Vladimir Somers, Victor Joos, Anthony Cioppa et al.

Tracking and identifying athletes on the pitch holds a central role in collecting essential insights from the game, such as estimating the total distance covered by players or understanding team tactics. This tracking and identification process is crucial for reconstructing the game state, defined by the athletes' positions and identities on a 2D top-view of the pitch, (i.e. a minimap). However, reconstructing the game state from videos captured by a single camera is challenging. It requires understanding the position of the athletes and the viewpoint of the camera to localize and identify players within the field. In this work, we formalize the task of Game State Reconstruction and introduce SoccerNet-GSR, a novel Game State Reconstruction dataset focusing on football videos. SoccerNet-GSR is composed of 200 video sequences of 30 seconds, annotated with 9.37 million line points for pitch localization and camera calibration, as well as over 2.36 million athlete positions on the pitch with their respective role, team, and jersey number. Furthermore, we introduce GS-HOTA, a novel metric to evaluate game state reconstruction methods. Finally, we propose and release an end-to-end baseline for game state reconstruction, bootstrapping the research on this task. Our experiments show that GSR is a challenging novel task, which opens the field for future research. Our dataset and codebase are publicly available at https://github.com/SoccerNet/sn-gamestate.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

Autonomous rendezvous and proximity operations around uncooperative, unknown spacecraft are critical for active debris removal and on-orbit servicing missions. A key component of such operations is the offline reconstruction of a 3D model of the target from a set of 2D images. This task is challenging due to two main factors. First, in-orbit illumination conditions exhibit considerable variability, and change rapidly over time. Second, the inaccuracy of pose information in the images, results in 3D reconstruction uncertainty. To overcome these challenges, we propose to extend Neural Radiance Fields with per-image degrees of freedom: a learnable appearance embedding that captures the illumination conditions specific to each image, and an image-specific pose correction term that refines its noisy pose label to increase 3D consistency across images. These parameters add minimal complexity, as they are learned jointly with the NeRF, yet they substantially improve robustness to illumination variability and pose inaccuracies. We validate our approach on three image sets representative of in-orbit operations, demonstrating its effectiveness for offline reconstruction and highlighting its suitability for online reconstruction, an open problem in the field.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

Spacecraft pose estimation networks require tens of thousands of CAD-rendered images to be trained. This reliance on synthetic CAD data (i) limits applicability to targets with reliable geometry prior, excluding uncooperative or poorly documented spacecraft, and (ii) causes poor generalization to real on-orbit conditions due to unrealistic illumination and material appearance. This paper introduces a NeRF-based image augmentation method that enables the learning of spacecraft pose estimators from only a few tens to a few hundreds of images. The method learns a Neural Radiance Field of the target and generates a large, diverse dataset through geometrically-consistent viewpoint and appearance augmentation. This augmented dataset enables the training of accurate target-specific pose estimators without requiring a CAD model or large synthetic datasets. Experiments show that our approach supports the training of accurate pose estimators from only 25 to 400 realistic images, even under severe illumination variations. When applied on large CAD-based synthetic datasets, the NeRF-based augmentation also enhances out-of-domain generalization, yielding improved robustness to real on-orbit conditions.

Maxime Zanella, Clément Fuchs, Christophe De Vleeschouwer et al.

The zero-shot capabilities of Vision-Language Models (VLMs) have been widely leveraged to improve predictive performance. However, previous works on transductive or test-time adaptation (TTA) often make strong assumptions about the data distribution, such as the presence of all classes. Our work challenges these favorable deployment scenarios, and introduces a more realistic evaluation framework, including: (i) a variable number of effective classes for adaptation within a single batch, and (ii) non-i.i.d. batches of test samples in online adaptation settings. We provide comprehensive evaluations, comparisons, and ablation studies that demonstrate how current transductive or TTA methods for VLMs systematically compromise the models' initial zero-shot robustness across various realistic scenarios, favoring performance gains under advantageous assumptions about the test samples' distributions. Furthermore, we introduce StatA, a versatile method that could handle a wide range of deployment scenarios, including those with a variable number of effective classes at test time. Our approach incorporates a novel regularization term designed specifically for VLMs, which acts as a statistical anchor preserving the initial text-encoder knowledge, particularly in low-data regimes. Code available at https://github.com/MaxZanella/StatA.

Clément Fuchs, Maxime Zanella, Christophe De Vleeschouwer

Online test-time adaptation (OTTA) of vision-language models (VLMs) has recently garnered increased attention to take advantage of data observed along a stream to improve future predictions. Unfortunately, existing methods rely on dataset-specific hyperparameters, significantly limiting their adaptability to unseen tasks. In response, we propose Online Gaussian Adaptation (OGA), a novel method that models the likelihoods of visual features using Gaussian distributions and incorporates zero-shot priors into an interpretable Maximum A Posteriori (MAP) estimation framework with fixed hyper-parameters across all datasets. We demonstrate that OGA outperforms state-of-the-art methods on most datasets and runs. Additionally, we show that combining OTTA with popular few-shot techniques (a practical yet overlooked setting in prior research) is highly beneficial. Furthermore, our experimental study reveals that common OTTA evaluation protocols, which average performance over at most three runs per dataset, are inadequate due to the substantial variability observed across runs for all OTTA methods. Therefore, we advocate for more rigorous evaluation practices, including increasing the number of runs and considering additional quantitative metrics, such as our proposed Expected Tail Accuracy (ETA), calculated as the average accuracy in the worst 10% of runs. We hope these contributions will encourage more rigorous and diverse evaluation practices in the OTTA community. Code is available at https://github.com/cfuchs2023/OGA .

Vladimir Somers, Baptiste Standaert, Victor Joos et al.

Online multi-object tracking has been recently dominated by tracking-by-detection (TbD) methods, where recent advances rely on increasingly sophisticated heuristics for tracklet representation, feature fusion, and multi-stage matching. The key strength of TbD lies in its modular design, enabling the integration of specialized off-the-shelf models like motion predictors and re-identification. However, the extensive usage of human-crafted rules for temporal associations makes these methods inherently limited in their ability to capture the complex interplay between various tracking cues. In this work, we introduce CAMEL, a novel association module for Context-Aware Multi-Cue ExpLoitation, that learns resilient association strategies directly from data, breaking free from hand-crafted heuristics while maintaining TbD's valuable modularity. At its core, CAMEL employs two transformer-based modules and relies on a novel association-centric training scheme to effectively model the complex interactions between tracked targets and their various association cues. Unlike end-to-end detection-by-tracking approaches, our method remains lightweight and fast to train while being able to leverage external off-the-shelf models. Our proposed online tracking pipeline, CAMELTrack, achieves state-of-the-art performance on multiple tracking benchmarks. Our code is available at https://github.com/TrackingLaboratory/CAMELTrack.

Maxime Zanella, Clément Fuchs, Ismail Ben Ayed et al.

Recent advances in few-shot adaptation for Vision-Language Models (VLMs) have greatly expanded their ability to generalize across tasks using only a few labeled examples. However, existing approaches primarily build upon the strong zero-shot priors of these models by leveraging carefully designed, task-specific prompts. This dependence on predefined class names can restrict their applicability, especially in scenarios where exact class names are unavailable or difficult to specify. To address this limitation, we introduce vocabulary-free few-shot learning for VLMs, a setting where target class instances - that is, images - are available but their corresponding names are not. We propose Similarity Mapping (SiM), a simple yet effective baseline that classifies target instances solely based on similarity scores with a set of generic prompts (textual or visual), eliminating the need for carefully handcrafted prompts. Although conceptually straightforward, SiM demonstrates strong performance, operates with high computational efficiency (learning the mapping typically takes less than one second), and provides interpretability by linking target classes to generic prompts. We believe that our approach could serve as an important baseline for future research in vocabulary-free few-shot learning. Code is available at https://github.com/MaxZanella/vocabulary-free-FSL.

Karim El Khoury, Maxime Zanella, Christophe De Vleeschouwer et al.

Remote Sensing Vision-Language Models (RSVLMs) have shown remarkable potential thanks to large-scale pretraining, achieving strong zero-shot performance on various tasks. However, their ability to generalize in low-data regimes, such as few-shot learning, remains insufficiently explored. In this work, we present the first structured benchmark for evaluating few-shot adaptation methods on RSVLMs. We conduct comprehensive experiments across ten remote sensing scene classification datasets, applying five widely used few-shot adaptation strategies to three state-of-the-art RSVLMs with varying backbones. Our findings reveal that models with similar zero-shot performance can exhibit markedly different behavior under few-shot adaptation, with some RSVLMs being inherently more amenable to such adaptation than others. The variability of performance and the absence of a clear winner among existing methods highlight the need for the development of more robust methods for few-shot adaptation tailored to RS. To facilitate future research, we provide a reproducible benchmarking framework and open-source code to systematically evaluate RSVLMs under few-shot conditions. The source code is publicly available on Github: https://github.com/elkhouryk/fewshot_RSVLMs

Gabriel Van Zandycke, Christophe De Vleeschouwer

Ball 3D localization in team sports has various applications including automatic offside detection in soccer, or shot release localization in basketball. Today, this task is either resolved by using expensive multi-views setups, or by restricting the analysis to ballistic trajectories. In this work, we propose to address the task on a single image from a calibrated monocular camera by estimating ball diameter in pixels and use the knowledge of real ball diameter in meters. This approach is suitable for any game situation where the ball is (even partly) visible. To achieve this, we use a small neural network trained on image patches around candidates generated by a conventional ball detector. Besides predicting ball diameter, our network outputs the confidence of having a ball in the image patch. Validations on 3 basketball datasets reveals that our model gives remarkable predictions on ball 3D localization. In addition, through its confidence output, our model improves the detection rate by filtering the candidates produced by the detector. The contributions of this work are (i) the first model to address 3D ball localization on a single image, (ii) an effective method for ball 3D annotation from single calibrated images, (iii) a high quality 3D ball evaluation dataset annotated from a single viewpoint. In addition, the code to reproduce this research is be made freely available at https://github.com/gabriel-vanzandycke/deepsport.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

We address the estimation of the 6D pose of an unknown target spacecraft relative to a monocular camera, a key step towards the autonomous rendezvous and proximity operations required by future Active Debris Removal missions. We present a novel method that enables an "off-the-shelf" spacecraft pose estimator, which is supposed to known the target CAD model, to be applied on an unknown target. Our method relies on an in-the wild NeRF, i.e., a Neural Radiance Field that employs learnable appearance embeddings to represent varying illumination conditions found in natural scenes. We train the NeRF model using a sparse collection of images that depict the target, and in turn generate a large dataset that is diverse both in terms of viewpoint and illumination. This dataset is then used to train the pose estimation network. We validate our method on the Hardware-In-the-Loop images of SPEED+ that emulate lighting conditions close to those encountered on orbit. We demonstrate that our method successfully enables the training of an off-the-shelf spacecraft pose estimation network from a sparse set of images. Furthermore, we show that a network trained using our method performs similarly to a model trained on synthetic images generated using the CAD model of the target.

Antoine Maiorca, Seyed Abolfazl Ghasemzadeh, Thierry Ravet et al.

Virtual Reality (VR) applications have revolutionized user experiences by immersing individuals in interactive 3D environments. These environments find applications in numerous fields, including healthcare, education, or architecture. A significant aspect of VR is the inclusion of self-avatars, representing users within the virtual world, which enhances interaction and embodiment. However, generating lifelike full-body self-avatar animations remains challenging, particularly in consumer-grade VR systems, where lower-body tracking is often absent. One method to tackle this problem is by providing an external source of motion information that includes lower body information such as full Cartesian positions estimated from RGB(D) cameras. Nevertheless, the limitations of these systems are multiples: the desynchronization between the two motion sources and occlusions are examples of significant issues that hinder the implementations of such systems. In this paper, we aim to measure the impact on the reconstruction of the articulated self-avatar's full-body pose of (1) the latency between the VR motion features and estimated positions, (2) the data acquisition rate, (3) occlusions, and (4) the inaccuracy of the position estimation algorithm. In addition, we analyze the motion reconstruction errors using ground truth and 3D Cartesian coordinates estimated from \textit{YOLOv8} pose estimation. These analyzes show that the studied methods are significantly sensitive to any degradation tested, especially regarding the velocity reconstruction error.

Dani Manjah, Davide Cacciarelli, Christophe De Vleeschouwer et al.

We present a scalable framework designed to craft efficient lightweight models for video object detection utilizing self-training and knowledge distillation techniques. We scrutinize methodologies for the ideal selection of training images from video streams and the efficacy of model sharing across numerous cameras. By advocating for a camera clustering methodology, we aim to diminish the requisite number of models for training while augmenting the distillation dataset. The findings affirm that proper camera clustering notably amplifies the accuracy of distilled models, eclipsing the methodologies that employ distinct models for each camera or a universal model trained on the aggregate camera data.

Benoît Gérin, Anaïs Halin, Anthony Cioppa et al.

In the era of the Internet of Things (IoT), objects connect through a dynamic network, empowered by technologies like 5G, enabling real-time data sharing. However, smart objects, notably autonomous vehicles, face challenges in critical local computations due to limited resources. Lightweight AI models offer a solution but struggle with diverse data distributions. To address this limitation, we propose a novel Multi-Stream Cellular Test-Time Adaptation (MSC-TTA) setup where models adapt on the fly to a dynamic environment divided into cells. Then, we propose a real-time adaptive student-teacher method that leverages the multiple streams available in each cell to quickly adapt to changing data distributions. We validate our methodology in the context of autonomous vehicles navigating across cells defined based on location and weather conditions. To facilitate future benchmarking, we release a new multi-stream large-scale synthetic semantic segmentation dataset, called DADE, and show that our multi-stream approach outperforms a single-stream baseline. We believe that our work will open research opportunities in the IoT and 5G eras, offering solutions for real-time model adaptation.

Alexandre Englebert, Anne-Sophie Collin, Olivier Cornu et al.

This paper proposes leveraging vision-language pretraining on bone X-rays paired with French reports to address downstream tasks of interest on bone radiography. A practical processing pipeline is introduced to anonymize and process French medical reports. Pretraining then consists in the self-supervised alignment of visual and textual embedding spaces derived from deep model encoders. The resulting image encoder is then used to handle various downstream tasks, including quantification of osteoarthritis, estimation of bone age on pediatric wrists, bone fracture and anomaly detection. Our approach demonstrates competitive performance on downstream tasks, compared to alternatives requiring a significantly larger amount of human expert annotations. Our work stands as the first study to integrate French reports to shape the embedding space devoted to bone X-Rays representations, capitalizing on the large quantity of paired images and reports data available in an hospital. By relying on generic vision-laguage deep models in a language-specific scenario, it contributes to the deployement of vision models for wider healthcare applications.

Dani Manjah, Tim Bary, Benoît Gérin et al.

Edge camera-based systems are continuously expanding, facing ever-evolving environments that require regular model updates. In practice, complex teacher models are run on a central server to annotate data, which is then used to train smaller models tailored to the edge devices with limited computational power. This work explores how to select the most useful images for training to maximize model quality while keeping transmission costs low. Our work shows that, for a similar training load (i.e., iterations), a high-confidence stream-based strategy coupled with a diversity-based approach produces a high-quality model with minimal dataset queries.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

On-orbit operations require the estimation of the relative 6D pose, i.e., position and orientation, between a chaser spacecraft and its target. While data-driven spacecraft pose estimation methods have been developed, their adoption in real missions is hampered by the lack of understanding of their decision process. This paper presents a method to visualize the 3D visual cues on which a given pose estimator relies. For this purpose, we train a NeRF-based image generator using the gradients back-propagated through the pose estimation network. This enforces the generator to render the main 3D features exploited by the spacecraft pose estimation network. Experiments demonstrate that our method recovers the relevant 3D cues. Furthermore, they offer additional insights on the relationship between the pose estimation network supervision and its implicit representation of the target spacecraft.

Elias De Smijter, Renaud Detry, Christophe De Vleeschouwer

We present the first systematic comparison of implicit and explicit Novel View Synthesis methods for space-based 3D object reconstruction, evaluating the role of appearance embeddings. While embeddings improve photometric fidelity by modeling lighting variation, we show they do not translate into meaningful gains in geometric accuracy - a critical requirement for space robotics applications. Using the SPEED+ dataset, we compare K-Planes, Gaussian Splatting, and Convex Splatting, and demonstrate that embeddings primarily reduce the number of primitives needed for explicit methods rather than enhancing geometric fidelity. Moreover, convex splatting achieves more compact and clutter-free representations than Gaussian splatting, offering advantages for safety-critical applications such as interaction and collision avoidance. Our findings clarify the limits of appearance embeddings for geometry-centric tasks and highlight trade-offs between reconstruction quality and representation efficiency in space scenarios.

Silvio Giancola, Anthony Cioppa, Marc Gutiérrez-Pérez et al.

The SoccerNet 2025 Challenges mark the fifth annual edition of the SoccerNet open benchmarking effort, dedicated to advancing computer vision research in football video understanding. This year's challenges span four vision-based tasks: (1) Team Ball Action Spotting, focused on detecting ball-related actions in football broadcasts and assigning actions to teams; (2) Monocular Depth Estimation, targeting the recovery of scene geometry from single-camera broadcast clips through relative depth estimation for each pixel; (3) Multi-View Foul Recognition, requiring the analysis of multiple synchronized camera views to classify fouls and their severity; and (4) Game State Reconstruction, aimed at localizing and identifying all players from a broadcast video to reconstruct the game state on a 2D top-view of the field. Across all tasks, participants were provided with large-scale annotated datasets, unified evaluation protocols, and strong baselines as starting points. This report presents the results of each challenge, highlights the top-performing solutions, and provides insights into the progress made by the community. The SoccerNet Challenges continue to serve as a driving force for reproducible, open research at the intersection of computer vision, artificial intelligence, and sports. Detailed information about the tasks, challenges, and leaderboards can be found at https://www.soccer-net.org, with baselines and development kits available at https://github.com/SoccerNet.

Benoit Brummer, Christophe De Vleeschouwer

This paper introduces the Raw Natural Image Noise Dataset (RawNIND), a diverse collection of paired raw images designed to support the development of denoising models that generalize across sensors, image development workflows, and styles. Two denoising methods are proposed: one operates directly on raw Bayer data, leveraging computational efficiency, while the other processes linear RGB images for improved generalization to different sensors, with both preserving flexibility for subsequent development. Both methods outperform traditional approaches which rely on developed images. Additionally, the integration of denoising and compression at the raw data level significantly enhances rate-distortion performance and computational efficiency. These findings suggest a paradigm shift toward raw data workflows for efficient and flexible image processing.

Antoine Legrand, Renaud Detry, Christophe De Vleeschouwer

We address the problem of estimating the relative 6D pose, i.e., position and orientation, of a target spacecraft, from a monocular image, a key capability for future autonomous Rendezvous and Proximity Operations. Due to the difficulty of acquiring large sets of real images, spacecraft pose estimation networks are exclusively trained on synthetic ones. However, because those images do not capture the illumination conditions encountered in orbit, pose estimation networks face a domain gap problem, i.e., they do not generalize to real images. Our work introduces a method that bridges this domain gap. It relies on a novel, end-to-end, neural-based architecture as well as a novel learning strategy. This strategy improves the domain generalization abilities of the network through multi-task learning and aggressive data augmentation policies, thereby enforcing the network to learn domain-invariant features. We demonstrate that our method effectively closes the domain gap, achieving state-of-the-art accuracy on the widespread SPEED+ dataset. Finally, ablation studies assess the impact of key components of our method on its generalization abilities.

Amir M. Mansourian, Vladimir Somers, Christophe De Vleeschouwer et al.

Effective tracking and re-identification of players is essential for analyzing soccer videos. But, it is a challenging task due to the non-linear motion of players, the similarity in appearance of players from the same team, and frequent occlusions. Therefore, the ability to extract meaningful embeddings to represent players is crucial in developing an effective tracking and re-identification system. In this paper, a multi-purpose part-based person representation method, called PRTreID, is proposed that performs three tasks of role classification, team affiliation, and re-identification, simultaneously. In contrast to available literature, a single network is trained with multi-task supervision to solve all three tasks, jointly. The proposed joint method is computationally efficient due to the shared backbone. Also, the multi-task learning leads to richer and more discriminative representations, as demonstrated by both quantitative and qualitative results. To demonstrate the effectiveness of PRTreID, it is integrated with a state-of-the-art tracking method, using a part-based post-processing module to handle long-term tracking. The proposed tracking method outperforms all existing tracking methods on the challenging SoccerNet tracking dataset.

Sédrick Stassin, Alexandre Englebert, Géraldin Nanfack et al.

EXplainable Artificial Intelligence (XAI) aims to help users to grasp the reasoning behind the predictions of an Artificial Intelligence (AI) system. Many XAI approaches have emerged in recent years. Consequently, a subfield related to the evaluation of XAI methods has gained considerable attention, with the aim to determine which methods provide the best explanation using various approaches and criteria. However, the literature lacks a comparison of the evaluation metrics themselves, that one can use to evaluate XAI methods. This work aims to fill this gap by comparing 14 different metrics when applied to nine state-of-the-art XAI methods and three dummy methods (e.g., random saliency maps) used as references. Experimental results show which of these metrics produces highly correlated results, indicating potential redundancy. We also demonstrate the significant impact of varying the baseline hyperparameter on the evaluation metric values. Finally, we use dummy methods to assess the reliability of metrics in terms of ranking, pointing out their limitations.

Seyed Abolfazl Ghasemzadeh, Gabriel Van Zandycke, Maxime Istasse et al.

This paper presents a unified framework to (i) locate the ball, (ii) predict the pose, and (iii) segment the instance mask of players in team sports scenes. Those problems are of high interest in automated sports analytics, production, and broadcast. A common practice is to individually solve each problem by exploiting universal state-of-the-art models, \eg, Panoptic-DeepLab for player segmentation. In addition to the increased complexity resulting from the multiplication of single-task models, the use of the off-the-shelf models also impedes the performance due to the complexity and specificity of the team sports scenes, such as strong occlusion and motion blur. To circumvent those limitations, our paper proposes to train a single model that simultaneously predicts the ball and the player mask and pose by combining the part intensity fields and the spatial embeddings principles. Part intensity fields provide the ball and player location, as well as player joints location. Spatial embeddings are then exploited to associate player instance pixels to their respective player center, but also to group player joints into skeletons. We demonstrate the effectiveness of the proposed model on the DeepSport basketball dataset, achieving comparable performance to the SoA models addressing each individual task separately.

Benoit Brummer, Christophe De Vleeschouwer

Convolutional autoencoders are now at the forefront of image compression research. To improve their entropy coding, encoder output is typically analyzed with a second autoencoder to generate per-variable parametrized prior probability distributions. We instead propose a compression scheme that uses a single convolutional autoencoder and multiple learned prior distributions working as a competition of experts. Trained prior distributions are stored in a static table of cumulative distribution functions. During inference, this table is used by an entropy coder as a look-up-table to determine the best prior for each spatial location. Our method offers rate-distortion performance comparable to that obtained with a predicted parametrized prior with only a fraction of its entropy coding and decoding complexity.

Adrien Deliège, Maxime Istasse, Ashwani Kumar et al.

In the framework of convolutional neural networks, downsampling is often performed with an average-pooling, where all the activations are treated equally, or with a max-pooling operation that only retains an element with maximum activation while discarding the others. Both of these operations are restrictive and have previously been shown to be sub-optimal. To address this issue, a novel pooling scheme, named\emph{ ordinal pooling}, is introduced in this work. Ordinal pooling rearranges all the elements of a pooling region in a sequence and assigns a different weight to each element based upon its order in the sequence. These weights are used to compute the pooling operation as a weighted sum of the rearranged elements of the pooling region. They are learned via a standard gradient-based training, allowing to learn a behavior anywhere in the spectrum of average-pooling to max-pooling in a differentiable manner. Our experiments suggest that it is advantageous for the networks to perform different types of pooling operations within a pooling layer and that a hybrid behavior between average- and max-pooling is often beneficial. More importantly, they also demonstrate that ordinal pooling leads to consistent improvements in the accuracy over average- or max-pooling operations while speeding up the training and alleviating the issue of the choice of the pooling operations and activation functions to be used in the networks. In particular, ordinal pooling mainly helps on lightweight or quantized deep learning architectures, as typically considered e.g. for embedded applications.

Carbonnelle Simon, Christophe De Vleeschouwer

Several works have shown that the regularization mechanisms underlying deep neural networks' generalization performances are still poorly understood. In this paper, we hypothesize that deep neural networks are regularized through their ability to extract meaningful clusters among the samples of a class. This constitutes an implicit form of regularization, as no explicit training mechanisms or supervision target such behaviour. To support our hypothesis, we design four different measures of intraclass clustering, based on the neuron- and layer-level representations of the training data. We then show that these measures constitute accurate predictors of generalization performance across variations of a large set of hyperparameters (learning rate, batch size, optimizer, weight decay, dropout rate, data augmentation, network depth and width).

Anne-Sophie Collin, Christophe De Vleeschouwer

In industrial vision, the anomaly detection problem can be addressed with an autoencoder trained to map an arbitrary image, i.e. with or without any defect, to a clean image, i.e. without any defect. In this approach, anomaly detection relies conventionally on the reconstruction residual or, alternatively, on the reconstruction uncertainty. To improve the sharpness of the reconstruction, we consider an autoencoder architecture with skip connections. In the common scenario where only clean images are available for training, we propose to corrupt them with a synthetic noise model to prevent the convergence of the network towards the identity mapping, and introduce an original Stain noise model for that purpose. We show that this model favors the reconstruction of clean images from arbitrary real-world images, regardless of the actual defects appearance. In addition to demonstrating the relevance of our approach, our validation provides the first consistent assessment of reconstruction-based methods, by comparing their performance over the MVTec AD dataset, both for pixel- and image-wise anomaly detection.

Antoine Vanderschueren, Victor Joos, Christophe De Vleeschouwer

We propose a novel approach to localize a 3D object from the intensity and depth information images provided by a Time-of-Flight (ToF) sensor. Our method uses two CNNs. The first one uses raw depth and intensity images as input, to segment the floor pixels, from which the extrinsic parameters of the camera are estimated. The second CNN is in charge of segmenting the object-of-interest. As a main innovation, it exploits the calibration estimated from the prediction of the first CNN to represent the geometric depth information in a coordinate system that is attached to the ground, and is thus independent of the camera elevation. In practice, both the height of pixels with respect to the ground, and the orientation of normals to the point cloud are provided as input to the second CNN. Given the segmentation predicted by the second CNN, the object is localized based on point cloud alignment with a reference model. Our experiments demonstrate that our proposed two-step approach improves segmentation and localization accuracy by a significant margin compared to a conventional CNN architecture, ignoring calibration and height maps, but also compared to PointNet++.

Gabriel Van Zandycke, Christophe De Vleeschouwer

This paper considers the task of detecting the ball from a single viewpoint in the challenging but common case where the ball interacts frequently with players while being poorly contrasted with respect to the background. We propose a novel approach by formulating the problem as a segmentation task solved by an efficient CNN architecture. To take advantage of the ball dynamics, the network is fed with a pair of consecutive images. Our inference model can run in real time without the delay induced by a temporal analysis. We also show that test-time data augmentation allows for a significant increase the detection accuracy. As an additional contribution, we publicly release the dataset on which this work is based.

Benoit Brummer, Christophe De Vleeschouwer

Most current research in the domain of image compression focuses solely on achieving state of the art compression ratio, but that is not always usable in today's workflow due to the constraints on computing resources. Constant market requirements for a low-complexity image codec have led to the recent development and standardization of a lightweight image codec named JPEG XS. In this work we show that JPEG XS compression can be adapted to a specific given task and content, such as preserving visual quality on desktop content or maintaining high accuracy in neural network segmentation tasks, by optimizing its gain and priority parameters using the covariance matrix adaptation evolution strategy.

Maxime Istasse, Julien Moreau, Christophe De Vleeschouwer

Assigning team labels to players in a sport game is not a trivial task when no prior is known about the visual appearance of each team. Our work builds on a Convolutional Neural Network (CNN) to learn a descriptor, namely a pixel-wise embedding vector, that is similar for pixels depicting players from the same team, and dissimilar when pixels correspond to distinct teams. The advantage of this idea is that no per-game learning is needed, allowing efficient team discrimination as soon as the game starts. In principle, the approach follows the associative embedding framework introduced in arXiv:1611.05424 to differentiate instances of objects. Our work is however different in that it derives the embeddings from a lightweight segmentation network and, more fundamentally, because it considers the assignment of the same embedding to unconnected pixels, as required by pixels of distinct players from the same team. Excellent results, both in terms of team labelling accuracy and generalization to new games/arenas, have been achieved on panoramic views of a large variety of basketball games involving players interactions and occlusions. This makes our method a good candidate to integrate team separation in many CNN-based sport analytics pipelines.