NBA_Streaming: A Large-Scale Benchmark for Fine-Grained Basketball Commentary Generation in Continuous Streams

2026-08-10

AI Summary

This work addresses the challenge of generating fine-grained, factually accurate commentary in real time from continuous basketball video streams—a task hindered by the lack of joint modeling of event boundaries, player identities, action details, and their temporal relationships. To bridge this gap, the study extends basketball commentary generation to the live streaming setting, introducing NBA_Streaming, a new benchmark comprising 307 hours of game footage with 35,000 temporally aligned annotations. The authors propose a two-stage framework under causal constraints: the first stage performs online event localization via ball-centric semantic anchoring and completeness assessment, while the second stage synthesizes semantically coherent commentary by integrating scene context, identity, and action cues. The approach significantly outperforms existing baselines in timing precision, factual accuracy, and descriptive granularity, establishing a new paradigm for streaming sports video understanding.

Abstract

Live basketball commentary generation requires determining when an event is sufficiently observable and describing it before subsequent events unfold. However, existing methods are primarily designed for pre-segmented clips or complete videos, making them unsuitable for continuous streams. Existing datasets also provide limited supervision for player identities, fine-grained actions, event attributes, and coherent event chains, restricting the factual richness of generated commentary. To address these limitations, we introduce NBA_Streaming, a large-scale benchmark for online fine-grained basketball commentary generation. It contains 307 hours of basketball broadcasts and approximately 35K temporally aligned events, with annotations of event boundaries, player identities, fine-grained actions, event chains, and natural-language commentary. By moving from isolated clips to continuous streams, NBA_Streaming enables unified evaluation of event localization, response reliability, factual grounding, and commentary quality under causal constraints. We further propose a causal two-stage framework that combines completion-first localization with ball-centric semantic grounding, enabling the system to identify complete events from observed streams and organize scene, event, identity, and action cues for commentary generation. Extensive experiments reveal the difficulty of NBA_Streaming, where existing baselines struggle with online timing, factual grounding, and fine-grained description. Our framework consistently improves over strong alternatives, while the remaining gap highlights NBA_Streaming as a valuable benchmark for streaming sports video understanding and generation.

Problem

• live commentary generation
• continuous video streams
• fine-grained action recognition
• event localization
• factual grounding

Innovation

• streaming commentary generation
• fine-grained action recognition
• causal event localization
• ball-centric semantic grounding
• sports video understanding