Apple and Fudan University Unveil StreamBridge: An Edge-Side Video Large Language Model Framework for Real-Time AI Responses to Video Streams
2025-05-13 / Read about 0 minute
Author:小编   

Apple has joined forces with Fudan University to launch StreamBridge, an innovative edge-side video large language model framework aimed at enhancing AI's ability to comprehend live video streams. Leveraging memory buffer and epoch decay compression techniques, StreamBridge effectively tackles the complexities of multi-round real-time comprehension and proactive response. Furthermore, it incorporates a lightweight, independently activated model, supported by the Stream-IT dataset comprising 600,000 samples. According to test results, integrating StreamBridge with mainstream video large language models, like Qwen2-VL, significantly boosts their performance, even outperforming some proprietary models. This groundbreaking development presents a unique technical solution for real-time video stream analysis.