BoLD (Body Language Dataset)
收藏资源简介:
BoLD数据集是由宾夕法尼亚州立大学创建的,用于面部表情识别研究。该数据集包含3450个视频,总计224,258帧,视频选自公开的AVA数据集,主要关注面部表情而非全身或部分遮挡的身体。数据集的创建过程涉及使用MTCNN自动验证视频,确保视频中只有一个演员的面部被检测到,并且面部占据的区域超过一定阈值。BoLD数据集的应用领域主要集中在面部表情识别,旨在解决面部表情识别中的数据不足问题,尤其是在自然环境下的表情识别。
The BoLD dataset was developed by Pennsylvania State University for research on facial expression recognition. It consists of 3,450 videos totaling 224,258 frames, selected from the publicly available AVA dataset, with a primary focus on facial expressions rather than full-body or partially occluded bodily regions. The dataset construction process involved automatically validating videos using MTCNN to ensure that only one facial region of a single actor is detected in each video, and that the occupied facial area exceeds a specified threshold. The main application scope of the BoLD dataset centers on facial expression recognition tasks, aiming to address the issue of insufficient data for facial expression recognition, especially expression recognition in naturalistic environments.

- 1Noisy Student Training using Body Language Dataset Improves Facial Expression Recognition宾夕法尼亚州立大学 · 2021年



