Fabian Deuser

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.

Hao Li, Fabian Deuser, Wenping Yina et al.

Nature disasters play a key role in shaping human-urban infrastructure interactions. Effective and efficient response to natural disasters is essential for building resilience and a sustainable urban environment. Two types of information are usually the most necessary and difficult to gather in disaster response. The first information is about disaster damage perception, which shows how badly people think that urban infrastructure has been damaged. The second information is geolocation awareness, which means how people whereabouts are made available. In this paper, we proposed a novel disaster mapping framework, namely CVDisaster, aiming at simultaneously addressing geolocalization and damage perception estimation using cross-view Street-View Imagery (SVI) and Very High-Resolution satellite imagery. CVDisaster consists of two cross-view models, where CVDisaster-Geoloc refers to a cross-view geolocalization model based on a contrastive learning objective with a Siamese ConvNeXt image encoder, and CVDisaster-Est is a cross-view classification model based on a Couple Global Context Vision Transformer (CGCViT). Taking Hurricane IAN as a case study, we evaluate the CVDisaster framework by creating a novel cross-view dataset (CVIAN) and conducting extensive experiments. As a result, we show that CVDisaster can achieve highly competitive performance (over 80% for geolocalization and 75% for damage perception estimation) with even limited fine-tuning efforts, which largely motivates future cross-view models and applications within a broader GeoAI research community. The data and code are publicly available at: https://github.com/tum-bgd/CVDisaster.

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.

Robin-Nico Kampa, Fabian Deuser, Anna Bößendörfer et al.

Autonomous AI coding agents are becoming a core tool for ML practitioners in industry and research alike. Despite this growing adoption, no standardized benchmark exists to evaluate their ability to design, implement, and train models from scratch across diverse domains. We introduce **1GC-7RC** (*Single Graphic Card: Seven Research Challenges*), a benchmark comprising seven ML tasks spanning language modeling, image classification, semantic segmentation, graph learning, tabular prediction, time-series forecasting, and text classification. Each task provides a locked data-preparation and evaluation script together with a baseline training script; the agent may only modify the training code, has no access to pretrained weights (with one controlled exception for semantic segmentation), no internet access, and must complete each task within a task-specific wall-clock budget (40-120 minutes) on a single GPU. We evaluate seven coding agents: five proprietary (Claude Code with Sonnet 4.6, Opus 4.6, and Opus 4.7; Codex CLI with GPT 5.5; and OpenCode with Qwen 3.6+) and two open-source (OpenCode with Kimi K2.5, Kimi K2.6). Across 5 runs per agent-task pair, we report substantial performance differences that reveal varying levels of implicit ML knowledge, planning ability, and time-budget management. The benchmark, harness, and all evaluation artifacts are publicly available on GitHub at https://github.com/Strolchii/1GC-7RC-Benchmark to facilitate reproducible comparison of future agents. Because our benchmark design is modular, the benchmark can be extended to new tasks and domains, adapted to different GPU budgets, and used to study multi-agent settings, making it a flexible platform for future research on autonomous research agents.

Hao Li, Fabian Deuser, Wenping Yin et al.

As Earth's climate changes, it is impacting disasters and extreme weather events across the planet. Record-breaking heat waves, drenching rainfalls, extreme wildfires, and widespread flooding during hurricanes are all becoming more frequent and more intense. Rapid and efficient response to disaster events is essential for climate resilience and sustainability. A key challenge in disaster response is to accurately and quickly identify disaster locations to support decision-making and resources allocation. In this paper, we propose a Probabilistic Cross-view Geolocalization approach, called ProbGLC, exploring new pathways towards generative location awareness for rapid disaster response. Herein, we combine probabilistic and deterministic geolocalization models into a unified framework to simultaneously enhance model explainability (via uncertainty quantification) and achieve state-of-the-art geolocalization performance. Designed for rapid diaster response, the ProbGLC is able to address cross-view geolocalization across multiple disaster events as well as to offer unique features of probabilistic distribution and localizability score. To evaluate the ProbGLC, we conduct extensive experiments on two cross-view disaster datasets (i.e., MultiIAN and SAGAINDisaster), consisting diverse cross-view imagery pairs of multiple disaster types (e.g., hurricanes, wildfires, floods, to tornadoes). Preliminary results confirms the superior geolocalization accuracy (i.e., 0.86 in Acc@1km and 0.97 in Acc@25km) and model explainability (i.e., via probabilistic distributions and localizability scores) of the proposed ProbGLC approach, highlighting the great potential of leveraging generative cross-view approach to facilitate location awareness for better and faster disaster response. The data and code is publicly available at https://github.com/bobleegogogo/ProbGLC

Fabian Deuser, Konrad Habel, Norbert Oswald

Cross-View Geo-Localisation is still a challenging task where additional modules, specific pre-processing or zooming strategies are necessary to determine accurate positions of images. Since different views have different geometries, pre-processing like polar transformation helps to merge them. However, this results in distorted images which then have to be rectified. Adding hard negatives to the training batch could improve the overall performance but with the default loss functions in geo-localisation it is difficult to include them. In this article, we present a simplified but effective architecture based on contrastive learning with symmetric InfoNCE loss that outperforms current state-of-the-art results. Our framework consists of a narrow training pipeline that eliminates the need of using aggregation modules, avoids further pre-processing steps and even increases the generalisation capability of the model to unknown regions. We introduce two types of sampling strategies for hard negatives. The first explicitly exploits geographically neighboring locations to provide a good starting point. The second leverages the visual similarity between the image embeddings in order to mine hard negative samples. Our work shows excellent performance on common cross-view datasets like CVUSA, CVACT, University-1652 and VIGOR. A comparison between cross-area and same-area settings demonstrate the good generalisation capability of our model.

Hannah Schieber, Fabian Deuser, Bernhard Egger et al.

Novel view synthesis using neural radiance fields (NeRF) is the state-of-the-art technique for generating high-quality images from novel viewpoints. Existing methods require a priori knowledge about extrinsic and intrinsic camera parameters. This limits their applicability to synthetic scenes, or real-world scenarios with the necessity of a preprocessing step. Current research on the joint optimization of camera parameters and NeRF focuses on refining noisy extrinsic camera parameters and often relies on the preprocessing of intrinsic camera parameters. Further approaches are limited to cover only one single camera intrinsic. To address these limitations, we propose a novel end-to-end trainable approach called NeRFtrinsic Four. We utilize Gaussian Fourier features to estimate extrinsic camera parameters and dynamically predict varying intrinsic camera parameters through the supervision of the projection error. Our approach outperforms existing joint optimization methods on LLFF and BLEFF. In addition to these existing datasets, we introduce a new dataset called iFF with varying intrinsic camera parameters. NeRFtrinsic Four is a step forward in joint optimization NeRF-based view synthesis and enables more realistic and flexible rendering in real-world scenarios with varying camera parameters.

Fabian Deuser, Konrad Habel, Martin Werner et al.

Retrieving relevant multimedia content is one of the main problems in a world that is increasingly data-driven. With the proliferation of drones, high quality aerial footage is now available to a wide audience for the first time. Integrating this footage into applications can enable GPS-less geo-localisation or location correction. In this paper, we present an orientation-guided training framework for UAV-view geo-localisation. Through hierarchical localisation orientations of the UAV images are estimated in relation to the satellite imagery. We propose a lightweight prediction module for these pseudo labels which predicts the orientation between the different views based on the contrastive learned embeddings. We experimentally demonstrate that this prediction supports the training and outperforms previous approaches. The extracted pseudo-labels also enable aligned rotation of the satellite image as augmentation to further strengthen the generalisation. During inference, we no longer need this orientation module, which means that no additional computations are required. We achieve state-of-the-art results on both the University-1652 and University-160k datasets.

Konrad Habel, Fabian Deuser, Norbert Oswald

Sports analytics benefits from recent advances in machine learning providing a competitive advantage for teams or individuals. One important task in this context is the performance measurement of individual players to provide reports and log files for subsequent analysis. During sport events like basketball, this involves the re-identification of players during a match either from multiple camera viewpoints or from a single camera viewpoint at different times. In this work, we investigate whether it is possible to transfer the out-standing zero-shot performance of pre-trained CLIP models to the domain of player re-identification. For this purpose we reformulate the contrastive language-to-image pre-training approach from CLIP to a contrastive image-to-image training approach using the InfoNCE loss as training objective. Unlike previous work, our approach is entirely class-agnostic and benefits from large-scale pre-training. With a fine-tuned CLIP ViT-L/14 model we achieve 98.44 % mAP on the MMSports 2022 Player Re-Identification challenge. Furthermore we show that the CLIP Vision Transformers have already strong OCR capabilities to identify useful player features like shirt numbers in a zero-shot manner without any fine-tuning on the dataset. By applying the Score-CAM algorithm we visualise the most important image regions that our fine-tuned model identifies when calculating the similarity score between two images of a player.

Fabian Deuser, Konrad Habel, Philipp J. Rösch et al.

Current architectures for multi-modality tasks such as visual question answering suffer from their high complexity. As a result, these architectures are difficult to train and require high computational resources. To address these problems we present a CLIP-based architecture that does not require any fine-tuning of the feature extractors. A simple linear classifier is used on the concatenated features of the image and text encoder. During training an auxiliary loss is added which operates on the answer types. The resulting classification is then used as an attention gate on the answer class selection. On the VizWiz 2022 Visual Question Answering Challenge we achieve 60.15 % accuracy on Task 1: Predict Answer to a Visual Question and AP score of 83.78 % on Task 2: Predict Answerability of a Visual Question.

Jiawei Ge, Xintian Zhang, Jiuxin Cao et al.

Cross-view Referring Multi-Object Tracking (CRMOT) aims to track multiple objects specified by natural language across multiple camera views, with globally consistent identities. Despite recent progress, existing methods rely heavily on costly frame-level spatial annotations and cross-view identity supervision. To reduce such reliance, we explore CRMOT under weak supervision by leveraging the capabilities of foundation models. However, our empirical study shows that directly applying foundation models such as SAM2 and SAM3, even with task-specific modifications, fails to accurately understand referring expressions and maintain consistent identities across views. Yet, they remain effective at producing reliable object tracklets that can serve as pseudo supervision. We therefore repurpose foundation models as pseudo-label generators and propose a two-stage framework for weakly supervised CRMOT, using only object category labels as coarse-grained supervision. In the first stage, we design an Affinity-guided Cross-view Re-prompting strategy to refine and associate SAM3-generated tracklets across cameras, producing reliable cross-view pseudo labels for subsequent training. In the second stage, we introduce ViewSAM, a CRMOT model built upon SAM2 that explicitly models view-aware cross-modal semantics. By formulating view-induced variations as learnable conditions, ViewSAM bridges the gap between view-variant visual observations and view-invariant textual expressions, enabling robust cross-view referring tracking with only approximately 10% additional parameters. Extensive experiments demonstrate that ViewSAM achieves SOTA performance under weak supervision and remains competitive with fully supervised methods.

Tobias Hallmen, Fabian Deuser, Norbert Oswald et al.

In this research, we introduce a novel methodology for assessing Emotional Mimicry Intensity (EMI) as part of the 6th Workshop and Competition on Affective Behavior Analysis in-the-wild. Our methodology utilises the Wav2Vec 2.0 architecture, which has been pre-trained on an extensive podcast dataset, to capture a wide array of audio features that include both linguistic and paralinguistic components. We refine our feature extraction process by employing a fusion technique that combines individual features with a global mean vector, thereby embedding a broader contextual understanding into our analysis. A key aspect of our approach is the multi-task fusion strategy that not only leverages these features but also incorporates a pre-trained Valence-Arousal-Dominance (VAD) model. This integration is designed to refine emotion intensity prediction by concurrently processing multiple emotional dimensions, thereby embedding a richer contextual understanding into our framework. For the temporal analysis of audio data, our feature fusion process utilises a Long Short-Term Memory (LSTM) network. This approach, which relies solely on the provided audio data, shows marked advancements over the existing baseline, offering a more comprehensive understanding of emotional mimicry in naturalistic settings, achieving the second place in the EMI challenge.

Tobias Hallmen, Robin-Nico Kampa, Fabian Deuser et al.

In this study, we present our methodology for two tasks: the Emotional Mimicry Intensity (EMI) Estimation Challenge and the Behavioural Ambivalence/Hesitancy (BAH) Recognition Challenge, both conducted as part of the 8th Workshop and Competition on Affective & Behavior Analysis in-the-wild. We utilize a Wav2Vec 2.0 model pre-trained on a large podcast dataset to extract various audio features, capturing both linguistic and paralinguistic information. Our approach incorporates a valence-arousal-dominance (VAD) module derived from Wav2Vec 2.0, a BERT text encoder, and a vision transformer (ViT) with predictions subsequently processed through a long short-term memory (LSTM) architecture or a convolution-like method for temporal modeling. We integrate the textual and visual modality into our analysis, recognizing that semantic content provides valuable contextual cues and underscoring that the meaning of speech often conveys more critical insights than its acoustic counterpart alone. Fusing in the vision modality helps in some cases to interpret the textual modality more precisely. This combined approach results in significant performance improvements, achieving in EMI $ρ_{\text{TEST}} = 0.706$ and in BAH $F1_{\text{TEST}} = 0.702$, securing first place in the EMI challenge and second place in the BAH challenge.

Boyi Chen, Zhangyu Wang, Fabian Deuser et al.

Accurate and robust image-based geo-localization at a global scale is challenging due to diverse environments, visually ambiguous scenes, and the lack of distinctive landmarks in many regions. While contrastive learning methods show promising performance by aligning features between street-view images and corresponding locations, they neglect the underlying spatial dependency in the geographic space. As a result, they fail to address the issue of false negatives -- image pairs that are both visually and geographically similar but labeled as negatives, and struggle to effectively distinguish hard negatives, which are visually similar but geographically distant. To address this issue, we propose a novel spatially regularized contrastive learning strategy that integrates a semivariogram, which is a geostatistical tool for modeling how spatial correlation changes with distance. We fit the semivariogram by relating the distance of images in feature space to their geographical distance, capturing the expected visual content in a spatial correlation. With the fitted semivariogram, we define the expected visual dissimilarity at a given spatial distance as reference to identify hard negatives and false negatives. We integrate this strategy into GeoCLIP and evaluate it on the OSV5M dataset, demonstrating that explicitly modeling spatial priors improves image-based geo-localization performance, particularly at finer granularity.

Robin-Nico Kampa, Fabian Deuser, Konrad Habel et al.

Plant phenotyping involves analyzing observable characteristics of plants to better understand their growth, health, and development. In the context of deep learning, this analysis is often approached through single-view classification or regression models. However, these methods often fail to capture all information required for accurate estimation of target phenotypic traits, which can adversely affect plant health assessment and harvest readiness prediction. To address this, the Growth Modelling (GroMo) Grand Challenge at ACM Multimedia 2025 provides a multi-view dataset featuring multiple plants and two tasks: Plant Age Prediction and Leaf Count Estimation. Each plant is photographed from multiple heights and angles, leading to significant overlap and redundancy in the captured information. To learn view-invariant embeddings, we incorporate 24 views, referred to as the selection vector, in a random selection. Our ViewSparsifier approach won both tasks. For further improvement and as a direction for future research, we also experimented with randomized view selection across all five height levels (120 views total), referred to as selection matrices.

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.

Johannes Flotzinger, Fabian Deuser, Achref Jaziri et al.

Adequate bridge inspection is increasingly challenging in many countries due to growing ailing stocks, compounded with a lack of staff and financial resources. Automating the key task of visual bridge inspection, classification of defects and building components on pixel level, improves efficiency, increases accuracy and enhances safety in the inspection process and resulting building assessment. Models overtaking this task must cope with an assortment of real-world conditions. They must be robust to variations in image quality, as well as background texture, as defects often appear on surfaces of diverse texture and degree of weathering. dacl10k is the largest and most diverse dataset for real-world concrete bridge inspections. However, the dataset exhibits class imbalance, which leads to notably poor model performance particularly when segmenting fine-grained classes such as cracks and cavities. This work introduces "synth-dacl", a compilation of three novel dataset extensions based on synthetic concrete textures. These extensions are designed to balance class distribution in dacl10k and enhance model performance, especially for crack and cavity segmentation. When incorporating the synth-dacl extensions, we observe substantial improvements in model robustness across 15 perturbed test sets. Notably, on the perturbed test set, a model trained on dacl10k combined with all synthetic extensions achieves a 2% increase in mean IoU, F1 score, Recall, and Precision compared to the same model trained solely on dacl10k.

Fabian Deuser, Philipp Hausenblas, Hannah Schieber et al.

Contrastive learning is a representational learning paradigm in which a neural network maps data elements to feature vectors. It improves the feature space by forming lots with an anchor and examples that are either positive or negative based on class similarity. Hard negative examples, which are close to the anchor in the feature space but from a different class, improve learning performance. Finding such examples of high quality efficiently in large, high-dimensional datasets is computationally challenging. In this paper, we propose a GPU-friendly Locality-Sensitive Hashing (LSH) scheme that quantizes real-valued feature vectors into binary representations for approximate nearest neighbor search. We investigate its theoretical properties and evaluate it on several datasets from textual and visual domain. Our approach achieves comparable or better performance while requiring significantly less computation than existing hard negative mining strategies.