In this episode we dive into Rex-Omni, a next-point prediction model that unifies detection, keypointing, and visual prompting; FineQuest, an “agent-of-thoughts” system for question-answering and reasoning over sports videos; and BroadTrack, a new framework for precise broadcast-camera tracking in soccer. Together, they reveal how cutting-edge AI is bridging the gap between pixels, play, and performance.

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:

Rex-Omni Demo
FineQuest: Adaptive Knowledge-Assisted Sports Video Understanding via Agent-of-Thoughts Reasoning
BroadTrack: Broadcast Camera Tracking for Soccer

In this of the SportsAI Podcast, we dive into three research papers that define how AI sees, understands, and improves sports performance and story telling.

• ExpertAF (CVPR 2025): AI that gives actionable video feedback, acting like a digital coach that helps athletes improve technique.

• SoccerHigh (ArXiv 2025): A dataset connecting video, commentary, and stats to measure soccer’s emotional and tactical “highs.”

• Efficient Framework for Sport Actions Identification (Springer 2025): A real-time fitness video analysis model using MediaPipe Pose to recognize 16 exercises and provide instant feedback.

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:

ExpertAF: Expert Actionable Feedback from Video
SoccerHigh: A Benchmark Dataset for Automatic Soccer Video Summarization
Efficient framework for sport actions identification and analysis in real-time fitness videos using neural network

In this episode, we talk about three research papers at the intersection of sports technology and computer vision. We explore the evolution of player tracking with a look at the SoccerTrack dataset, see how the SportsSloMo benchmark is creating stunning slow-motion replays, and finally discuss how AI is being used for fine-grained skill estimation in basketball with the Basket dataset

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:

SoccerTrack
Team Track
SoccerTrack V2
SoccerTrack Challenge
SportsSloMo: A New Benchmark and Baselines for Human-centric Video Frame Interpolation
BASKET : A Large-Scale Video Dataset for Fine-Grained Skill Estimation

In this special episode, we dive into the latest groundbreaking research presented at CVPR 2025, the premier conference in computer vision. We explore fascinating papers transforming sports analysis, including:

• Ball spin and trajectory prediction in table tennis broadcasts.

• Advanced puck localization techniques revolutionizing ice hockey analytics.

• AthletePose3D, an innovative benchmark dataset pushing the limits of 3D human pose estimation.

• VGG-T, winner of the Best Paper Award, introducing the first feed-forward structure-from-motion method enhanced by ConvexSplatting.

• UniSoccer, a universal AI model aiming to understand soccer footage from every angle.

• FSBench, the first-of-its-kind benchmark dataset focused on artistic and technical understanding of figure skating performances.

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:

Conferences: CVPR, ICCV, ECCV
VGGT: Visual Geometry Grounded Transformer
Towards Universal Soccer Video Understanding
FSBench: Figure Skating Benchmark for Advancing Artistic Sports Understanding
AthletePose3D: A Benchmark Dataset for 3D Human Pose Estimation and Kinematic Validation in Athletic Movements
Ice Hockey Puck Localization Using Contextual Cues
Towards Ball Spin and Trajectory Analysis in Table Tennis Broadcast Videos via Physically Grounded Synthetic-to-Real Transfer

In this episode, we talk with Ian, a researcher and active referee, about how AI is reshaping the world of officiating. We explore innovations in referee decision support, new VAR-like systems, and the role of computer vision in improving accuracy and consistency. A look into the future of fairer, more efficient refereeing at all levels of sport. 

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:

VARS: Video Assistant Referee System for Automated Soccer Decision Making from Multiple Views
X-VARS: Introducing Explainability in Football Refereeing with Multi-Modal Large Language Model

In this episode, we dive deep into the world of sports video data extraction. We break down the various types of sports data, from traditional box scores and detailed event (play-by-play) data to advanced tracking metrics and explore how each type contributes uniquely to sports analytics. The discussion highlights the technological challenges involved in extracting accurate data, including issues around precision, automation, and scalability. Finally, we share our insights and conclusions on the state of sports data extraction, and what we consider to be a key dependency. 

Your hosts:
Alexandre H. Bustamante

Anthony Cioppa

Silvio Giancola

References:
SoccerNet-v2: A Dataset and Benchmarks for Holistic Understanding of Broadcast Soccer Videos

References:
SoccerNet-v2: A Dataset and Benchmarks for Holistic Understanding of Broadcast Soccer Videos
SoccerNet Game State Reconstruction: End-to-End Athlete Tracking and Identification on a Minimap
X-VARS: Introducing Explainability in Football Refereeing with Multi-Modal Large Language Models
SoccerNet-Caption: Dense Video Captioning for Soccer Broadcasts Commentaries