InfiniBench
收藏资源简介:
InfiniBench是一个用于长期视频理解的新数据集,由富士通研究, 富士通有限公司创建。该数据集旨在评估和提升长期视频理解模型的性能,特别是针对长视频序列的处理能力。数据集的创建过程结合了视觉压缩技术和问题自适应机制,以减少内存需求并提高模型效率。该数据集主要应用于视频问答和长期视频理解任务,旨在解决现有方法在处理长视频时面临的内存和性能瓶颈问题。
InfiniBench is a novel dataset for long-term video understanding, developed by Fujitsu Research, Fujitsu Limited. This dataset aims to evaluate and enhance the performance of long-term video understanding models, with a particular focus on their capability to process lengthy video sequences. The construction of this dataset integrates visual compression techniques and question-adaptive mechanisms to reduce memory requirements and improve model efficiency. It is primarily applied to video question answering and long-term video understanding tasks, and is designed to address the memory and performance bottlenecks faced by existing methods when handling long videos.




