Continuous Perception Benchmark
收藏资源简介:
Continuous Perception Benchmark数据集由斯坦福大学创建,旨在推动视频理解模型的连续感知能力。该数据集包含200个视频实例,每个视频时长20秒,帧率为30fps,涵盖10个不同的物体类别。数据集通过模拟环境OmniGibson生成,要求模型连续处理视频以准确计数特定物体,模拟人类连续视觉处理的过程。该数据集主要用于评估和促进模型在复杂视频内容中进行精确物体识别和计数的能力。
The Continuous Perception Benchmark dataset was created by Stanford University, aiming to advance the continuous perception capabilities of video understanding models. This dataset includes 200 video instances, each with a duration of 20 seconds and a frame rate of 30 fps, covering 10 distinct object categories. Generated using the OmniGibson simulation environment, the dataset requires models to process videos continuously to accurately count specific objects, simulating the process of human continuous visual processing. It is primarily used to evaluate and improve the ability of models to perform precise object recognition and counting in complex video content.

- 1Continuous Perception Benchmark斯坦福大学 · 2024年



