Waveplot-based Dataset for Multi-class Human Action Analysis

Published: 25 July 2023| Version 1 | DOI: 10.17632/3vsz7v53pn.1


This dataset comprises an assortment of waveplot images representing diverse human actions. Waveplot images are time-amplitude representations of audio signals that encapsulate the variation of audio amplitude over time. In this dataset, the audio signals correspond to disparate human actions, such as walking, running, jumping, and dancing. The waveplot images are created by plotting the amplitude of the audio signals against time, with each image representing a segment of the audio signal. The dataset is explicitly designed for tasks like human action recognition, classification, segmentation, and detection based on auditory cues. It serves as a valuable resource for training and evaluating machine learning models that analyze human actions predicated on audio signals. The dataset caters well to researchers and practitioners in the disciplines of signal processing, computer vision, and machine learning, who are keen on devising algorithms for human action analysis using audio signals. Crucially, the dataset is annotated with labels that denote the type of human action represented in each waveplot image. This ensures a supervised learning environment conducive for the development and testing of prediction models.


Steps to reproduce



Edith Cowan University, University of Western Australia


Computer Vision Representation, Benchmarking, Multimodality, Image Analysis, Action Recognition