DHF1K
收藏资源简介:
DHF1K是一个大规模的视频注视点预测数据集,由北京智能信息技术实验室创建。该数据集包含1000个高质量视频序列,覆盖广泛的场景、运动、物体类型和背景复杂度。DHF1K旨在通过提供多样化和具有挑战性的动态场景,推动视频注视点建模的发展。数据集中的每个视频都由17名观察者进行注视点标注,总帧数超过600,000帧。DHF1K不仅用于视频注视点预测模型的训练和评估,还提供了丰富的视频类别和属性标注,以深入理解动态场景自由观看中的注视引导机制。
DHF1K is a large-scale video gaze prediction dataset developed by the Beijing Intelligent Information Technology Laboratory. This dataset contains 1000 high-quality video sequences covering a wide spectrum of scenarios, motions, object types and background complexities. DHF1K aims to advance the development of video gaze prediction modeling by providing diverse and challenging dynamic scenes. Each video in the dataset is annotated by 17 observers, with the total number of frames exceeding 600,000. Besides being used for training and evaluating video gaze prediction models, DHF1K also provides rich video category and attribute annotations to facilitate in-depth understanding of the gaze guidance mechanism during free viewing of dynamic scenes.

- 1Revisiting Video Saliency: A Large-scale Benchmark and a New Model北京智能信息技术实验室 · 2018年



