SensorData for Human Activity Recognition (HAR)

This repository contains pre-processed sensor datasets based on the original datasets from SSL-Wearables. Please refer to the SSL-Wearables repository for the original data sources (here).

Dataset Description

The collection includes several widely used datasets in the field of ubiquitous computing and wearable sensing. These datasets have been pre-processed to align with the experimental settings of recent state-of-the-art methods.

• Repository: sha-ce/SensorData
• Paper: (paper url)
• Point of Contact: (Name/Email)

Supported Tasks

• Human Activity Recognition (HAR): Classification of human activities based on wearable sensor data (accelerometer).
• Text-to-Signal Generation: Evaluating generative models on sensor data.

Included Datasets

Based on the benchmarks, this repository includes data from the following sources:

1. ADL (MotionSense/MobiAct): Activities of Daily Living recorded from smartphone sensors.
2. Opportunity: Activities of daily living recorded in a sensor-rich environment, focusing on gestures and object interactions.
3. PAMAP2: Physical Activity Monitoring dataset containing data from 9 subjects performing 18 different physical activities.
4. RealWorld (HAR): Acceleration data collected from smartphones/wearables in realistic settings.
5. WISDM: Wireless Sensor Data Mining dataset containing accelerometer data for activity recognition.

Dataset Structure

Data Format

The data is likely stored in .npy, or .npz formats.

• Input Features: Multi-channel time-series data (e.g., 3-axis accelerometer).
• Labels: Integer class labels corresponding to specific activities (e.g., Walking, Running, Sitting).

Usage

You can download the files directly or use the Hugging Face huggingface_hub library to download specific datasets.

from huggingface_hub import hf_hub_download
import numpy as np

# Example: Downloading PAMAP2 data
file_path = hf_hub_download(repo_id="sha-ce/SensorData", filename="pamap2.npz") # Adjust filename
data = np.load(file_path)

print(data.files)
# Output might be: ['x_train', 'y_train', 'x_test', 'y_test']

or

hf download sha-ce/SensorData --repo-type dataset --local-dir .