Human Action Recognition by Learning Bases of Action Attributes and Parts Bangpeng Yao, Xiaoye Jiang, Aditya Khosla, Andy Lai Lin, Leonidas Guibas, and Li Fei-Fei Stanford University
Outline Introduction Action Bases Learning the Dual-Sparse Action Bases and Reconstruction Coefficients Experiments
Introduction Human action recognition in still images Contributions A general image classification problem Human-object interaction Parts + Attributes Contributions Represent each image by using a sparse set of action bases that are meaningful to the content of the image Effectively learn these bases given far-from-perfect detections of action attributes and parts without meticulous human labeling
Action Bases Attributes and parts Attributes: verb, learned by discriminative classifiers Parts: object parts and poselets, learned by pre-trained object detectors and poselet detectors A vector of the normalized confidence scores obtained from these classifiers and detectors is used to represent this image.
Action Bases High-order interactions of image attributes and parts is used to represent each image and SVMs are trained for action classification
Dual-sparsity Learning
Experiments PASCAL actions Stanford 40 actions
PASCAL
Stanford 40 actions