FacialFlowNet (FFN)
收藏资源简介:
FacialFlowNet (FFN) 是由复旦大学智能信息处理重点实验室创建的一个大规模面部光流数据集,包含9,635个身份和105,970对图像。该数据集旨在提供前所未有的多样性,以支持详细的面部和头部运动分析。数据集通过使用Blender和Cycles渲染引擎生成合成帧和光流标签,涵盖了多种面部表情和头部姿态。创建过程包括3D面部重建、UV纹理提取和数据集渲染,确保了数据的高质量和真实性。该数据集主要应用于面部光流估计和分解,旨在提高面部表情分析的准确性,特别是在微表情识别领域。
FacialFlowNet (FFN) is a large-scale facial optical flow dataset developed by the Key Laboratory of Smart Information Processing, Fudan University. It consists of 9,635 distinct human identities and 105,970 image pairs. This dataset aims to provide unprecedented diversity to support detailed facial and head motion analysis. The dataset is generated using Blender and Cycles rendering engines to produce synthetic frames and optical flow labels, covering a wide range of facial expressions and head poses. Its creation pipeline includes 3D facial reconstruction, UV texture extraction and dataset rendering, which ensures high data quality and authenticity. This dataset is mainly applied to facial optical flow estimation and decomposition, with the objective of enhancing the accuracy of facial expression analysis, especially in the domain of micro-expression recognition.




