NBA_Streaming: A Large-Scale Benchmark for Fine-Grained Basketball Commentary Generation in Continuous Streams
2026-08-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created NBA_Streaming, a large new dataset to help computers provide live basketball commentary from continuous video streams, instead of just pre-cut clips. This dataset includes detailed info like when events happen, who the players are, their actions, and related commentary. They also developed a two-step method that first detects events and then uses ball-focused cues to generate more accurate commentary. Their tests show that this is a hard task and their method improves over existing ones, but challenges remain. Overall, their work sets a new standard for understanding and describing live sports videos.
live commentary generationevent localizationplayer identificationfine-grained actionscausal inferencebasketball video analysissemantic groundingstreaming videonatural language generationbenchmark dataset
Authors
Lifang Wu, Yuyang Wu, Yangdong Gao, Fengyu Liu, Ya Jing, Liang Wang
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.