pith. sign in

arxiv: 2512.11016 · v2 · submitted 2025-12-11 · 💻 cs.CV · cs.AI

SoccerMaster: A Vision Foundation Model for Soccer Understanding

Haolin Yang , Jiayuan Rao , Haoning Wu , Weidi Xie This is my paper

Pith reviewed 2026-05-16 23:05 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords soccer understandingvision foundation modelmulti-task pretrainingsports video analysisathlete detectionevent classificationautomated data curationSoccerFactory
0
0 comments X

The pith

A single soccer-specific vision foundation model unifies detection, identification, and event reasoning while outperforming separate expert models on each task.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents SoccerMaster as the first vision foundation model built specifically for soccer, trained through supervised multi-task pretraining on a mix of existing video datasets. It introduces SoccerFactory, an automated pipeline that generates large-scale spatial annotations without manual labeling, to support this unified training. The model is shown to handle both low-level perception tasks such as athlete detection and high-level reasoning tasks such as event classification within one network. Evaluations indicate that this single model surpasses task-specific expert models across the tested downstream tasks. The result suggests that domain-focused multi-task pretraining can reduce the need for separate specialized systems in sports video analysis.

Core claim

SoccerMaster is a unified vision foundation model that performs supervised multi-task pretraining on soccer data curated by the SoccerFactory pipeline and integrated public datasets; this single model then outperforms dedicated task-specific expert models on a range of downstream soccer visual understanding tasks that span fine-grained perception such as athlete detection and identification to high-level semantic reasoning such as event classification.

What carries the argument

SoccerMaster, a vision foundation model trained via supervised multi-task pretraining on soccer-specific spatial annotations generated by the automated SoccerFactory pipeline.

If this is right

  • One pretrained network can replace multiple separate models for detection, identification, and event classification in soccer footage.
  • Automated spatial annotation pipelines can supply the volume of labeled data needed for multi-task pretraining without proportional manual effort.
  • Integrating several existing soccer datasets yields a richer pretraining resource than any single dataset alone.
  • The same architecture demonstrates measurable gains on both low-level perception and high-level reasoning tasks after the shared pretraining stage.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same data-curation approach could be adapted to create foundation models for other team sports that share similar visual structure and event semantics.
  • A deployed SoccerMaster could reduce engineering overhead in broadcast analysis systems by serving multiple query types from one model checkpoint.
  • Extending the pretraining to include temporal video clips rather than single frames might further improve performance on action and event tasks that depend on motion.

Load-bearing premise

The SoccerFactory automated annotation pipeline produces labels of high enough quality and without systematic biases that would prevent effective multi-task pretraining or downstream generalization.

What would settle it

A controlled evaluation on a held-out soccer video dataset where SoccerMaster achieves lower accuracy than the best task-specific expert model on at least two of the reported downstream tasks.

Figures

Figures reproduced from arXiv: 2512.11016 by Haolin Yang, Haoning Wu, Jiayuan Rao, Weidi Xie.

Figure 1
Figure 1. Figure 1: SoccerMaster is a unified soccer-specific vision foundation model that leverages diverse soccer content, including images and videos, to support a wide range of soccer understanding tasks, such as commentary generation, detection, tracking, classification, etc. Abstract Soccer understanding has recently garnered growing re￾search interest due to its domain-specific complexity and unique challenges. Unlike … view at source ↗
Figure 2
Figure 2. Figure 2: Automated Data Curation Pipeline. Our pipeline processes input videos through three stages: (i) field registration establishes geometric correspondences between image and canonical pitch coordinates via keypoint detection; (ii) tracking and identification trans￾forms frames into athlete trajectories through detection, role and team classification, and ReID-based tracking; and (iii) post-processing refineme… view at source ↗
Figure 3
Figure 3. Figure 3: SoccerMaster Architecture. (a) The architecture of SoccerMaster, which encodes both soccer videos and images through spatial and temporal attention modules to generate semantically rich representations. (b) The pretraining tasks and downstream adaptations of SoccerMaster across both spatial perception and semantic understanding tasks. SigLIP 2 [62], and obtain the final semantic features, de￾noted as Fsem … view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative Results of our Automatic Curation Pipeline. Comparison between our predictions (left) and ground truth annota￾tions (right) on the SoccerNet-GSR test set. Our pipeline demonstrates robust performance across diverse scenarios. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top-view Pitch Visualization of Pipeline Results. Athlete positions are mapped to standardized pitch coordinates via estimated camera parameters. Each row is organized as: input image (left), our predictions (middle), and ground truth annotations (right). Athletes are color-coded by role: referees (orange, labeled “RE”), left team (red), and right team (blue). Non-referee athletes are labeled with arbitrar… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Results of SoccerMaster. SoccerMaster can simultaneously execute multiple soccer understanding tasks on a video clip, including athlete detection, pitch registration, multiple object tracking, event classification, and commentary generation. Frames are arranged in temporal order from left to right and top to bottom. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
read the original abstract

Soccer understanding has recently garnered growing research interest due to its domain-specific complexity and unique challenges. Unlike prior works that typically rely on isolated, task-specific expert models, this work aims to propose a unified model to handle diverse soccer visual understanding tasks, ranging from fine-grained perception (e.g., athlete detection and identification) to high-level semantic reasoning (e.g., event classification). Concretely, our contributions are threefold: (i) we present SoccerMaster, the first soccer-specific vision foundation model that unifies diverse tasks within a single framework via supervised multi-task pretraining; (ii) we develop an automated data curation pipeline, SoccerFactory, to generate scalable spatial annotations, and integrate multiple existing soccer video datasets as a comprehensive pretraining data resource for multi-task pretraining; and (iii) we conduct extensive evaluations demonstrating that SoccerMaster consistently outperforms task-specific expert models across diverse downstream tasks, highlighting its breadth and superiority. The data, code, and model will be publicly available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces SoccerMaster, the first soccer-specific vision foundation model that unifies diverse tasks (athlete detection/identification, event classification, etc.) via supervised multi-task pretraining. It contributes an automated SoccerFactory pipeline to curate scalable spatial annotations by integrating existing soccer video datasets, and reports extensive evaluations showing consistent outperformance over task-specific expert models on downstream tasks. The data, code, and model are promised to be released publicly.

Significance. If the outperformance claims hold after rigorous validation of annotation quality and ablations, the work would provide a useful unified baseline for soccer video understanding, demonstrating the value of multi-task pretraining in a specialized domain and potentially reducing reliance on separate expert models for perception and reasoning tasks.

major comments (3)
  1. [§3] §3 (SoccerFactory pipeline): the description of generating 'scalable spatial annotations' by integrating existing datasets provides no quantitative validation such as precision/recall on held-out frames, human agreement scores, or error typology for detection, jersey identification, or event labels. This is load-bearing for the central claim because any systematic biases (e.g., under-detection of occluded players) could artifactually drive the reported downstream gains.
  2. [Experiments section] Experiments section and results tables: the manuscript asserts 'consistent outperformance' and 'breadth and superiority' but the summary provides no detailed quantitative tables, ablation studies on task weighting or data scale, or error analysis. Without these, attribution of gains specifically to the unified multi-task approach versus data volume or task selection remains unverifiable.
  3. [§4] §4 (Evaluation protocol): potential data leakage or inconsistent train/test splits between the SoccerFactory pretraining corpus and downstream benchmarks is not addressed, nor are details on whether task-specific baselines were trained with equivalent data volume and augmentation. This directly affects the fairness of the superiority claims.
minor comments (2)
  1. [Abstract] Abstract: specify the exact number of tasks, datasets, and downstream benchmarks to allow readers to assess the scope of the 'extensive evaluations' claim.
  2. [Methods] Notation and methods: provide the explicit multi-task loss formulation and how per-task weights are set or learned, as this is needed for reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript to incorporate the suggested clarifications and additions, which will strengthen the presentation of SoccerFactory, the experimental results, and the evaluation protocol.

read point-by-point responses

  1. Referee: [§3] §3 (SoccerFactory pipeline): the description of generating 'scalable spatial annotations' by integrating existing datasets provides no quantitative validation such as precision/recall on held-out frames, human agreement scores, or error typology for detection, jersey identification, or event labels. This is load-bearing for the central claim because any systematic biases (e.g., under-detection of occluded players) could artifactually drive the reported downstream gains.

    Authors: We agree that quantitative validation of the automated annotations is necessary to substantiate the pipeline's reliability. In the revised manuscript we will add a dedicated validation subsection reporting precision/recall on held-out frames, inter-annotator agreement scores on a sampled subset, and a categorized error analysis covering detection misses, jersey mis-identifications, and event label inaccuracies. These metrics will be computed against manual ground truth and will explicitly discuss potential biases such as occlusion handling. revision: yes

  2. Referee: [Experiments section] Experiments section and results tables: the manuscript asserts 'consistent outperformance' and 'breadth and superiority' but the summary provides no detailed quantitative tables, ablation studies on task weighting or data scale, or error analysis. Without these, attribution of gains specifically to the unified multi-task approach versus data volume or task selection remains unverifiable.

    Authors: We will expand the Experiments section with full per-task quantitative tables (including all baselines), new ablation studies varying task weights and pretraining data scale, and a systematic error analysis. These additions will allow readers to isolate the contribution of multi-task pretraining from data volume effects and will be placed in the main text or a comprehensive appendix. revision: yes

  3. Referee: [§4] §4 (Evaluation protocol): potential data leakage or inconsistent train/test splits between the SoccerFactory pretraining corpus and downstream benchmarks is not addressed, nor are details on whether task-specific baselines were trained with equivalent data volume and augmentation. This directly affects the fairness of the superiority claims.

    Authors: We will add an explicit evaluation-protocol subsection that documents the exact train/test splits, confirms temporal and video-level separation between the SoccerFactory pretraining corpus and all downstream benchmarks to preclude leakage, and states that every task-specific baseline was re-trained using the same data volume and augmentation pipeline as SoccerMaster. These details will be summarized in a table for transparency. revision: yes

Circularity Check

0 steps flagged

No circularity: standard empirical multi-task pretraining with independent evaluation

full rationale

The paper's derivation consists of curating data via SoccerFactory, performing supervised multi-task pretraining, and reporting empirical outperformance on downstream tasks. No equations, fitted parameters, or self-referential definitions appear in the provided text. The central claim reduces to measured accuracy on held-out evaluations rather than any quantity defined by construction from the inputs. No self-citations, uniqueness theorems, or ansatzes are invoked as load-bearing steps. This is a conventional empirical ML contribution whose validity hinges on data quality and experimental controls, not on internal circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the effectiveness of supervised multi-task pretraining for cross-task generalization and on the assumption that SoccerFactory produces reliable spatial annotations at scale; both are standard domain assumptions in computer vision rather than new axioms or invented entities.

axioms (1)
  • domain assumption Supervised multi-task pretraining on combined soccer datasets improves performance on individual downstream tasks compared with single-task training
    Invoked in the description of the pretraining stage and the claim of outperformance

pith-pipeline@v0.9.0 · 5471 in / 1234 out tokens · 30481 ms · 2026-05-16T23:05:32.783006+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. SoccerLens: Grounded Soccer Video Understanding Beyond Accuracy

    cs.CV 2026-05 unverdicted novelty 7.0

    SoccerLens benchmark shows state-of-the-art soccer VLMs achieve strong classification accuracy yet fail to exceed 50% grounding performance on annotated visual cues and underutilize temporal information.

  2. SoccerLens: Grounded Soccer Video Understanding Beyond Accuracy

    cs.CV 2026-05 unverdicted novelty 7.0

    SoccerLens benchmark shows state-of-the-art soccer VLMs achieve high classification accuracy yet fail to exceed 50% visual grounding performance and underutilize temporal information.

  3. Towards Temporal Compositional Reasoning in Long-Form Sports Videos

    cs.CV 2026-04 unverdicted novelty 7.0

    SportsTime benchmark and CoTR method improve multimodal AI's temporal compositional reasoning and evidence grounding in long-form sports videos.

Reference graph

Works this paper leans on

90 extracted references · 90 canonical work pages · cited by 2 Pith papers · 3 internal anchors

  1. [1]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report.arXiv preprint arXiv:2502.13923, 2025. 3

    work page Pith review Pith/arXiv arXiv 2025

  2. [2]

    Jersey number recognition using keyframe identification from low-resolution broadcast videos

    Bavesh Balaji, Jerrin Bright, Harish Prakash, Yuhao Chen, David A Clausi, and John Zelek. Jersey number recognition using keyframe identification from low-resolution broadcast videos. InProceedings of the 6th International Workshop on Multimedia Content Analysis in Sports, 2023. 1, 2

    work page 2023

  3. [3]

    Meteor: An automatic metric for mt evaluation with improved correlation with hu- man judgments

    Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with hu- man judgments. InProceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, 2005. 9

    work page 2005

  4. [4]

    Evaluating mul- tiple object tracking performance: the clear mot metrics

    Keni Bernardin and Rainer Stiefelhagen. Evaluating mul- tiple object tracking performance: the clear mot metrics. EURASIP Journal on Image and Video Processing, 2008. 8

    work page 2008

  5. [5]

    Is space-time attention all you need for video understanding? InProceedings of the International Conference on Machine Learning, 2021

    Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? InProceedings of the International Conference on Machine Learning, 2021. 5

    work page 2021

  6. [6]

    Observation-centric sort: Rethinking sort for robust multi-object tracking

    Jinkun Cao, Jiangmiao Pang, Xinshuo Weng, Rawal Khirod- kar, and Kris Kitani. Observation-centric sort: Rethinking sort for robust multi-object tracking. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2023. 8

    work page 2023

  7. [7]

    Camera calibration and player local- ization in soccernet-v2 and investigation of their representa- tions for action spotting

    Anthony Cioppa, Adrien Deliege, Floriane Magera, Sil- vio Giancola, Olivier Barnich, Bernard Ghanem, and Marc Van Droogenbroeck. Camera calibration and player local- ization in soccernet-v2 and investigation of their representa- tions for action spotting. InProceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition Work- shops, 2021. 1

    work page 2021

  8. [8]

    Scaling up soccer- net with multi-view spatial localization and re-identification

    Anthony Cioppa, Adrien Deliege, Silvio Giancola, Bernard Ghanem, and Marc Van Droogenbroeck. Scaling up soccer- net with multi-view spatial localization and re-identification. Scientific Data, 2022. 1, 15

    work page 2022

  9. [9]

    Soccernet-tracking: Multiple object tracking dataset and benchmark in soccer videos

    Anthony Cioppa, Silvio Giancola, Adrien Deliege, Le Kang, Xin Zhou, Zhiyu Cheng, Bernard Ghanem, and Marc Van Droogenbroeck. Soccernet-tracking: Multiple object tracking dataset and benchmark in soccer videos. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2022. 1, 2, 8

    work page 2022

  10. [10]

    ArXiv abs/2409.10587(2024), https://api.semanticscholar.org/CorpusID:272693834

    Anthony Cioppa, Silvio Giancola, Vladimir Somers, Vic- tor Joos, Floriane Magera, Jan Held, Seyed Abol- fazl Ghasemzadeh, Xin Zhou, Karolina Seweryn, Mateusz Kowalczyk, et al. Soccernet 2024 challenges results.arXiv preprint arXiv:2409.10587, 2024. 2

    work page arXiv 2024

  11. [11]

    Sportsmot: A large multi- object tracking dataset in multiple sports scenes

    Yutao Cui, Chenkai Zeng, Xiaoyu Zhao, Yichun Yang, Gangshan Wu, and Limin Wang. Sportsmot: A large multi- object tracking dataset in multiple sports scenes. InProceed- ings of the International Conference on Computer Vision,

    work page

  12. [12]

    Soccernet-v2: A dataset and benchmarks for holis- tic understanding of broadcast soccer videos

    Adrien Deliege, Anthony Cioppa, Silvio Giancola, Meisam J Seikavandi, Jacob V Dueholm, Kamal Nasrollahi, Bernard Ghanem, Thomas B Moeslund, and Marc Van Droogen- broeck. Soccernet-v2: A dataset and benchmarks for holis- tic understanding of broadcast soccer videos. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, 2021. 1,...

    work page 2021

  13. [13]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. InProceedings of the International Conference on Learning Representati...

    work page 2021

  14. [14]

    Strongsort: Make deep- sort great again.IEEE Transactions on Multimedia, 2023

    Yunhao Du, Zhicheng Zhao, Yang Song, Yanyun Zhao, Fei Su, Tao Gong, and Hongying Meng. Strongsort: Make deep- sort great again.IEEE Transactions on Multimedia, 2023. 3, 8

    work page 2023

  15. [15]

    Enhancing soccer camera calibration through keypoint exploitation

    Nikolay S Falaleev and Ruilong Chen. Enhancing soccer camera calibration through keypoint exploitation. InACM Multimedia Workshops, 2024. 2, 3

    work page 2024

  16. [16]

    Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 2025

    Roya Firoozi, Johnathan Tucker, Stephen Tian, Anirudha Majumdar, Jiankai Sun, Weiyu Liu, Yuke Zhu, Shuran Song, Ashish Kapoor, Karol Hausman, et al. Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 2025. 2

    work page 2025

  17. [17]

    A survey for founda- tion models in autonomous driving

    Haoxiang Gao, Zhongruo Wang, Yaqian Li, Kaiwen Long, Ming Yang, and Yiqing Shen. A survey for foundation mod- els in autonomous driving.arXiv preprint arXiv:2402.01105,

    work page arXiv

  18. [18]

    Multiple object track- ing as id prediction

    Ruopeng Gao, Ji Qi, and Limin Wang. Multiple object track- ing as id prediction. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2025. 7, 8

    work page 2025

  19. [19]

    Soccernet: A scalable dataset for action spotting in soccer videos

    Silvio Giancola, Mohieddine Amine, Tarek Dghaily, and Bernard Ghanem. Soccernet: A scalable dataset for action spotting in soccer videos. InProceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition Work- shops, 2018. 1, 2

    work page 2018

  20. [20]

    Soccernet 2022 challenges results

    Silvio Giancola, Anthony Cioppa, Adrien Deli `ege, Floriane Magera, Vladimir Somers, Le Kang, Xin Zhou, Olivier Bar- nich, Christophe De Vleeschouwer, Alexandre Alahi, et al. Soccernet 2022 challenges results. InProceedings of the 5th International ACM Workshop on Multimedia Content Anal- ysis in Sports, 2022. 2, 8

    work page 2022

  21. [21]

    ArXiv abs/2508.19182(2025), https://api.semanticscholar.org/CorpusID:280870241

    Silvio Giancola, Anthony Cioppa, et al. Soccernet 2025 chal- lenges results.arXiv preprint arXiv:2508.19182, 2025. 2, 4

    work page arXiv 2025

  22. [22]

    From broadcast to min- imap: Achieving state-of-the-art soccernet game state recon- struction

    Vladimir Golovkin, Nikolay Nemtsev, Vasyl Shandyba, Oleg Udin, Nikita Kasatkin, Pavel Kononov, Anton Afanasiev, Sergey Ulasen, and Andrei Boiarov. From broadcast to min- imap: Achieving state-of-the-art soccernet game state recon- struction. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition Workshops, 2025. 4

    work page 2025

  23. [23]

    The Llama 3 Herd of Models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Ab- hinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783, 2024. 7

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  24. [24]

    Pnlcalib: Sports field registration via points and lines optimization.arXiv preprint arXiv:2404.08401, 2024

    Marc Guti ´errez-P´erez and Antonio Agudo. Pnlcalib: Sports field registration via points and lines optimization.arXiv preprint arXiv:2404.08401, 2024. 2, 3, 6, 7, 8, 15 10

    work page arXiv 2024

  25. [25]

    Vars: Video assistant referee system for automated soccer decision making from multiple views

    Jan Held, Anthony Cioppa, Silvio Giancola, Abdullah Hamdi, Bernard Ghanem, and Marc Van Droogenbroeck. Vars: Video assistant referee system for automated soccer decision making from multiple views. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion Workshops, 2023. 1, 2

    work page 2023

  26. [26]

    X-vars: Introducing explainability in football refereeing with multi- modal large language models

    Jan Held, Hani Itani, Anthony Cioppa, Silvio Giancola, Bernard Ghanem, and Marc Van Droogenbroeck. X-vars: Introducing explainability in football refereeing with multi- modal large language models. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2024. 1, 2

    work page 2024

  27. [27]

    Ibrahim, Srikanth Muralidharan, Zhiwei Deng, Arash Vahdat, and Greg Mori

    Mostafa S. Ibrahim, Srikanth Muralidharan, Zhiwei Deng, Arash Vahdat, and Greg Mori. A hierarchical deep temporal model for group activity recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2016. 2

    work page 2016

  28. [28]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma. Adam: A method for stochastic opti- mization.arXiv preprint arXiv:1412.6980, 2014. 17

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  29. [29]

    Maria Koshkina and James H. Elder. A general framework for jersey number recognition in sports video. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2024. 1, 4, 16

    work page 2024

  30. [30]

    Sports-qa: A large-scale video question answering bench- mark for complex and professional sports,

    Haopeng Li, Andong Deng, Qiuhong Ke, Jun Liu, Hos- sein Rahmani, Yulan Guo, Bernt Schiele, and Chen Chen. Sports-qa: A large-scale video question answering bench- mark for complex and professional sports.arXiv preprint arXiv:2401.01505, 2024. 2

    work page arXiv 2024

  31. [31]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. InProceed- ings of the International Conference on Machine Learning,

    work page

  32. [32]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InPro- ceedings of the International Conference on Machine Learn- ing, 2023. 2, 7

    work page 2023

  33. [33]

    Multisports: A multi-person video dataset of spatio-temporally localized sports actions

    Yixuan Li, Lei Chen, Runyu He, Zhenzhi Wang, Gangshan Wu, and Limin Wang. Multisports: A multi-person video dataset of spatio-temporally localized sports actions. InPro- ceedings of the International Conference on Computer Vi- sion, 2021. 2

    work page 2021

  34. [34]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText Summarization Branches Out, 2004. 9

    work page 2004

  35. [35]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ar. Focal loss for dense object detection. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017. 6

    work page 2017

  36. [36]

    F3set: Towards analyzing fast, frequent, and fine-grained events from videos

    Zhaoyu Liu, Kan Jiang, Murong Ma, Zhe Hou, Yun Lin, and Jin Song Dong. F3set: Towards analyzing fast, frequent, and fine-grained events from videos. InProceedings of the In- ternational Conference on Learning Representations, 2025. 2

    work page 2025

  37. [37]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InProceedings of the International Confer- ence on Learning Representations, 2019. 17

    work page 2019

  38. [38]

    Hota: A higher order metric for evaluating multi-object tracking.International Journal of Computer Vision, 2021

    Jonathon Luiten, Aljosa Osep, Patrick Dendorfer, Philip Torr, Andreas Geiger, Laura Leal-Taix´e, and Bastian Leibe. Hota: A higher order metric for evaluating multi-object tracking.International Journal of Computer Vision, 2021. 8

    work page 2021

  39. [39]

    Npu rgbd dataset and a feature-enhanced lstm-dgcn method for action recognition of basketball players+.Applied Sci- ences, 2021

    Chunyan Ma, Jizhuang Fan, Jing-Yue Yao, and Tao Zhang. Npu rgbd dataset and a feature-enhanced lstm-dgcn method for action recognition of basketball players+.Applied Sci- ences, 2021. 2

    work page 2021

  40. [40]

    A universal protocol to bench- mark camera calibration for sports

    Floriane Magera, Thomas Hoyoux, Olivier Barnich, and Marc Van Droogenbroeck. A universal protocol to bench- mark camera calibration for sports. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion Workshops, 2024. 1, 2, 8

    work page 2024

  41. [41]

    Broadtrack: Broadcast camera tracking for soccer

    Floriane Magera, Thomas Hoyoux, Olivier Barnich, and Marc Van Droogenbroeck. Broadtrack: Broadcast camera tracking for soccer. InWinter Conference on Applications of Computer Vision, 2025. 1

    work page 2025

  42. [42]

    Multi-task learning for joint re-identification, team affiliation, and role classifi- cation for sports visual tracking

    Amir M Mansourian, Vladimir Somers, Christophe De Vleeschouwer, and Shohreh Kasaei. Multi-task learning for joint re-identification, team affiliation, and role classifi- cation for sports visual tracking. InProceedings of the 6th International Workshop on Multimedia Content Analysis in Sports, 2023. 1, 2, 3, 8

    work page 2023

  43. [43]

    Leapfrog diffusion model for stochastic trajec- tory prediction

    Weibo Mao, Chenxin Xu, Qi Zhu, Siheng Chen, and Yan- feng Wang. Leapfrog diffusion model for stochastic trajec- tory prediction. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2023. 2

    work page 2023

  44. [44]

    Soccernet- caption: Dense video captioning for soccer broadcasts com- mentaries

    Hassan Mkhallati, Anthony Cioppa, Silvio Giancola, Bernard Ghanem, and Marc Van Droogenbroeck. Soccernet- caption: Dense video captioning for soccer broadcasts com- mentaries. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition Workshops, 2023. 1, 2, 16

    work page 2023

  45. [45]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. Transactions on Machine Learning Research, 2024. 2

    work page 2024

  46. [46]

    Basket: A large- scale video dataset for fine-grained skill estimation

    Yulu Pan, Ce Zhang, and Gedas Bertasius. Basket: A large- scale video dataset for fine-grained skill estimation. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2025. 2

    work page 2025

  47. [47]

    Bleu: a method for automatic evaluation of machine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. InAssociation for Computational Linguistics,

    work page

  48. [48]

    AJ Piergiovanni and Michael S. Ryoo. Fine-grained activ- ity recognition in baseball videos. InProceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition Workshops, 2018. 2

    work page 2018

  49. [49]

    Goal: A challenging knowledge-grounded video captioning bench- mark for real-time soccer commentary generation

    Ji Qi, Jifan Yu, Teng Tu, Kunyu Gao, Yifan Xu, Xinyu Guan, Xiaozhi Wang, Bin Xu, Lei Hou, Juanzi Li, et al. Goal: A challenging knowledge-grounded video captioning bench- mark for real-time soccer commentary generation. InPro- ceedings of the ACM International Conference on Informa- tion and Knowledge Management, 2023. 1, 2 11

    work page 2023

  50. [50]

    Sports video captioning via attentive motion representation and group relationship modeling.IEEE Transactions on Cir- cuits and Systems for Video Technology, 2020

    Mengshi Qi, Yunhong Wang, Annan Li, and Jiebo Luo. Sports video captioning via attentive motion representation and group relationship modeling.IEEE Transactions on Cir- cuits and Systems for Video Technology, 2020. 2

    work page 2020

  51. [51]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. InProceedings of the International Conference on Machine Learning, 2021. 2

    work page 2021

  52. [52]

    Matchtime: Towards automatic soccer game commentary generation

    Jiayuan Rao, Haoning Wu, Chang Liu, Yanfeng Wang, and Weidi Xie. Matchtime: Towards automatic soccer game commentary generation. InProceedings of the Confer- ence on Empirical Methods in Natural Language Processing,

    work page

  53. [53]

    1, 2, 3, 4, 5, 7, 8, 9, 15, 16, 17

    work page

  54. [54]

    Multi-agent system for comprehensive soccer understanding

    Jiayuan Rao, Zifeng Li, Haoning Wu, Ya Zhang, Yanfeng Wang, and Weidi Xie. Multi-agent system for comprehensive soccer understanding. InACM Multimedia, 2025. 1, 2

    work page 2025

  55. [55]

    Towards universal soccer video under- standing

    Jiayuan Rao, Haoning Wu, Hao Jiang, Ya Zhang, Yanfeng Wang, and Weidi Xie. Towards universal soccer video under- standing. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition, 2025. 1, 2, 3, 4, 5, 6, 7, 15, 16, 17

    work page 2025

  56. [56]

    Sam 2: Seg- ment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Seg- ment anything in images and videos. InProceedings of the International Conference on Learning Representations,

    work page

  57. [57]

    Performance measures and a data set for multi-target, multi-camera tracking

    Ergys Ristani, Francesco Solera, Roger Zou, Rita Cucchiara, and Carlo Tomasi. Performance measures and a data set for multi-target, multi-camera tracking. InProceedings of the European Conference on Computer Vision, 2016. 8

    work page 2016

  58. [58]

    Finegym: A hierarchical video dataset for fine-grained action understand- ing

    Dian Shao, Yue Zhao, Bo Dai, and Dahua Lin. Finegym: A hierarchical video dataset for fine-grained action understand- ing. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020. 2

    work page 2020

  59. [59]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Husz ´ar, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016. 6

    work page 2016

  60. [60]

    DINOv3

    Oriane Sim ´eoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamonjisoa, et al. Dinov3.arXiv preprint arXiv:2508.10104, 2025. 2, 7

    work page Pith review Pith/arXiv arXiv 2025

  61. [61]

    Soccernet game state reconstruction: End-to- end athlete tracking and identification on a minimap

    Vladimir Somers, Victor Joos, Anthony Cioppa, Silvio Gian- cola, Seyed Abolfazl Ghasemzadeh, Floriane Magera, Bap- tiste Standaert, Amir M Mansourian, Xin Zhou, Shohreh Kasaei, et al. Soccernet game state reconstruction: End-to- end athlete tracking and identification on a minimap. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Reco...

    work page 2024

  62. [62]

    Computer vision for sports: Current applications and research topics.Computer Vision and Image Understanding, 2017

    Graham Thomas, Rikke Gade, Thomas B Moeslund, Peter Carr, and Adrian Hilton. Computer vision for sports: Current applications and research topics.Computer Vision and Image Understanding, 2017. 2

    work page 2017

  63. [63]

    SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muham- mad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision-language en- coders with improved semantic understanding, localization, and dense features.arXiv preprint arXiv:2502.14786, 2025. 2, 3, 6, 7, 9, 17

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  64. [64]

    Semi-supervised training to im- prove player and ball detection in soccer

    Renaud Vandeghen, Anthony Cioppa, and Marc Van Droogenbroeck. Semi-supervised training to im- prove player and ball detection in soccer. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2022. 1, 2

    work page 2022

  65. [65]

    Yolov8: A novel object detection algorithm with enhanced performance and robust- ness

    Rejin Varghese and M Sambath. Yolov8: A novel object detection algorithm with enhanced performance and robust- ness. InInternational Conference on Advances in Data En- gineering and Intelligent Computing Systems, 2024. 3

    work page 2024

  66. [66]

    Cider: Consensus-based image description evalua- tion

    Ramakrishna Vedantam, C Lawrence Zitnick, and Devi Parikh. Cider: Consensus-based image description evalua- tion. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015. 9

    work page 2015

  67. [67]

    Vggt: Vi- sual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Vi- sual geometry grounded transformer. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2025. 2

    work page 2025

  68. [68]

    Tacticai: an ai assistant for football tactics.Nature Commu- nications, 2024

    Zhe Wang, Petar Veli ˇckovi´c, Daniel Hennes, Nenad Tomaˇsev, Laurel Prince, Michael Kaisers, Yoram Bachrach, Romuald Elie, Li Kevin Wenliang, Federico Piccinini, et al. Tacticai: an ai assistant for football tactics.Nature Commu- nications, 2024. 19

    work page 2024

  69. [69]

    Towards generalist foundation model for radiology by leveraging web-scale 2d&3d medical data

    Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Hui Hui, Yanfeng Wang, and Weidi Xie. Towards generalist foundation model for radiology by leveraging web-scale 2d&3d medical data. Nature Communications, 2025. 2

    work page 2025

  70. [70]

    Sports video analysis on large-scale data

    Dekun Wu, He Zhao, Xingce Bao, and Richard P Wildes. Sports video analysis on large-scale data. InProceedings of the European Conference on Computer Vision, 2022. 2

    work page 2022

  71. [71]

    A simple yet effective knowl- edge guided method for entity-aware video captioning on a basketball benchmark.Neurocomputing, 2025

    Zeyu Xi, Ge Shi, Xuefen Li, Junchi Yan, Zun Li, Lifang Wu, Zilin Liu, and Liang Wang. A simple yet effective knowl- edge guided method for entity-aware video captioning on a basketball benchmark.Neurocomputing, 2025

    work page 2025

  72. [72]

    Eika: Explicit & implicit knowledge-augmented network for entity-aware sports video captioning.Expert Systems with Applications, 2025

    Zeyu Xi, Ge Shi, Haoying Sun, Bowen Zhang, Shuyi Li, and Lifang Wu. Eika: Explicit & implicit knowledge-augmented network for entity-aware sports video captioning.Expert Systems with Applications, 2025

    work page 2025

  73. [73]

    Player-centric multimodal prompt generation for large lan- guage model based identity-aware basketball video caption- ing

    Zeyu Xi, Haoying Sun, Yaofei Wu, Junchi Yan, Haoran Zhang, Lifang Wu, Liang Wang, and Changwen Chen. Player-centric multimodal prompt generation for large lan- guage model based identity-aware basketball video caption- ing. InProceedings of the International Conference on Com- puter Vision, 2025. 2

    work page 2025

  74. [74]

    Sportqa: A benchmark for sports understanding in large language models

    Haotian Xia, Zhengbang Yang, Yuqing Wang, Rhys Tracy, Yun Zhao, Dongdong Huang, Zezhi Chen, Yan Zhu, Yuan- fang Wang, and Weining Shen. Sportqa: A benchmark for sports understanding in large language models. InProceed- 12 ings of the Conference of the North American Chapter of the Association for Computational Linguistics, 2024. 2

    work page 2024

  75. [75]

    Sportu: A comprehensive sports understanding benchmark for multimodal large language models

    Haotian Xia, Zhengbang Yang, Junbo Zou, Rhys Tracy, Yuqing Wang, Chi Lu, Christopher Lai, Yanjun He, Xun Shao, Zhuoqing Xie, et al. Sportu: A comprehensive sports understanding benchmark for multimodal large language models. InProceedings of the International Conference on Learning Representations, 2025. 2

    work page 2025

  76. [76]

    Language-guided audio-visual learn- ing for long-term sports assessment

    Huangbiao Xu, Xiao Ke, Huanqi Wu, Rui Xu, Yuezhou Li, and Wenzhong Guo. Language-guided audio-visual learn- ing for long-term sports assessment. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2025. 2

    work page 2025

  77. [77]

    Finediving: A fine-grained dataset for procedure-aware action quality assessment

    Jinglin Xu, Yongming Rao, Xumin Yu, Guangyi Chen, Jie Zhou, and Jiwen Lu. Finediving: A fine-grained dataset for procedure-aware action quality assessment. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, 2022. 2

    work page 2022

  78. [78]

    Depth any- thing v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiao- gang Xu, Jiashi Feng, and Hengshuang Zhao. Depth any- thing v2. InConference on Neural Information Processing Systems, 2024. 2

    work page 2024

  79. [79]

    Timesoccer: An end-to-end multimodal large language model for soccer commentary generation

    Ling You, Wenxuan Huang, Xinni Xie, Xiangyi Wei, Bangyan Li, Shaohui Lin, Yang Li, and Changbo Wang. Timesoccer: An end-to-end multimodal large language model for soccer commentary generation. InACM Multi- media, 2025. 2

    work page 2025

  80. [80]

    Scaling vision transformers

    Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lu- cas Beyer. Scaling vision transformers. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2022. 2, 5

    work page 2022

Showing first 80 references.