The paper introduces SlowFast networks for video recognition, which consist of a Slow pathway operating at a low frame rate to capture spatial semantics and a Fast pathway operating at a high frame rate to capture fine temporal resolution motion. The Fast pathway is designed to be lightweight by reducing its channel capacity while maintaining useful temporal information. The models achieve state-of-the-art performance on major video recognition benchmarks, including Kinetics, Charades, and AVA. The authors highlight the effectiveness of the SlowFast concept, which separates spatial structures and temporal events, and provide detailed experimental results and ablation studies to support their claims. The code for the models is available on GitHub.The paper introduces SlowFast networks for video recognition, which consist of a Slow pathway operating at a low frame rate to capture spatial semantics and a Fast pathway operating at a high frame rate to capture fine temporal resolution motion. The Fast pathway is designed to be lightweight by reducing its channel capacity while maintaining useful temporal information. The models achieve state-of-the-art performance on major video recognition benchmarks, including Kinetics, Charades, and AVA. The authors highlight the effectiveness of the SlowFast concept, which separates spatial structures and temporal events, and provide detailed experimental results and ablation studies to support their claims. The code for the models is available on GitHub.