HAR 15 Human Action Recognition Dataset

* This dataset supports online use.Click here to jump.

Dataset Introduction

The full name of this dataset is Human Action Recognition. It contains 15 different categories of human activities, about 12k+ labeled images (including verification images). Each image has only one human activity category and is saved in a separate folder of the labeled category.

Human action recognition (HAR) aims to understand human actions and assign labels to each action. It has a wide range of applications and thus has attracted increasing attention in the field of computer vision. Human actions can be represented using various data modalities, such as RGB, skeleton, depth, infrared, point cloud, event stream, audio, acceleration, radar, and WiFi signals, which encode useful but different information from different sources and have various advantages depending on the application scenario.

Dataset File

• train– Contains all the images used to train the model, a total of 15 folders, namely "Calling", "Clapping", "Riding a Bicycle", "Dancing", "Drinking", "Eating", "Fighting", "Hugging", "Laughing", "Listening to Music", "Running", "Sitting", "Sleeping", "Texting", "Using Laptop", which contain images of corresponding human activities.
• test– Contains 5400 images of human activities. For these images, predictions can be made based on the corresponding class names - "calling", "clapping", "biking", "dancing", "drinking", "eating", "fighting", "hugging", "laughing", "listening to music", "running", "sitting", "sleeping", "texting", "using laptop".
• Testing_set.csv – This is the order of predictions for each image submitted on the platform. Make sure you download the predictions and image file names in the same order as given in this file.
• sample_submission: This is a csv file containing sample commits for the data sprint.