Lightning Talk: Ball Tracking and Detection in Soccer Videos - Comparison of... Maciej Szymkowski

This talk focuses on ball tracking and detection in soccer videos, comparing Vision-Language Models (VLMs) with traditional pipelines. The speaker, Maciej Szymkowski from Future Processing, evaluates on-premise models like Qwen 3 and InternVL-3.5 against cloud-based solutions such as Gemini 3 and GPT-5, alongside a traditional YOLOv11-based pipeline. Utilizing soccer match recordings from varying environmental conditions, the presentation details the methodologies used, including manual annotation for a ground truth database and the fine-tuning of YOLOv11. The effectiveness of each approach is assessed through traditional machine learning metrics such as accuracy, precision, and recall, offering insights on the best practices for ball detection and tracking.