Physically Plausible Data Augmentations for Wearable IMU-based HAR

This repository contains the implementation of Physically Plausible Data Augmentation (PPDA) for wearable Inertial Measurement Unit (IMU)-based Human Activity Recognition (HAR)

📄 Read our paper (arXiv)

---

Figure: PPDAs introduce physically plausible variations in wearable IMU data using physics simulation, compared to conventional STDAs which directly apply signal transformations.

📌 Introduction

This repository provides the official implementation of Physically Plausible Data Augmentation (PPDA) for wearable IMU-based Human Activity Recognition (HAR). PPDA uses the WIMUSim physics simulation framework to generate realistic variations in:

• Movement Amplitude
• Movement Speed
• Sensor Placement
• Hardware-related Effects (noise and bias)

These augmentations help improve model generalization and reduce the need for large-scale labeled data collection, compared to traditional Signal Transformation-based Data Augmentations (STDAs).

⚙️ Installation

Prerequisites

• Python 3.10 or higher
• Git

1. Clone this repository and WIMUSim repository

git clone https://github.com/USERNAME/PPDA.git
git clone https://github.com/STRCWearlab/WIMUSim.git
cd PPDA

2. Install dependencies

% 
pip install -e WIMUSim
pip install -r requirements.txt

3. Configure environment variables Set the WANDB_ENTITY environment variable to your Weights & Biases entity name.

WANDB_ENTITY=`replace_with_your_wandb_entity`

3. Prepare datasets Download datasets (REALDISP, REALWORLD, MM-Fit) as described in data/README.md.

🧪 Running Scripts

Individual Augmentation Assessment (Paper Section IV.B)

Run the following command to assess individual augmentations (e.g. MM-Fit dataset):

# Movement amplitude scaling
python scripts/mmfit_ppda_indiv.py --magscale --magscale_sigma 0.2
# Movement amplitude warping
python scripts/mmfit_ppda_indiv.py --magwarp --magwarp_sigma 0.2 --magwarp_knot 4
# Movement speed scaling
python scripts/mmfit_ppda_indiv.py --timescale --timescale_scale_min 0.8 --timescale_scale_max 1.2
# Movement speed warping
python scripts/mmfit_ppda_indiv.py --timewarp --timewarp_sigma 0.1 --timewarp_knot 4
# Sensor rotation/placement variation
python scripts/mmfit_ppda_indiv.py --rotation --rotation_range_x -25,25 --rotation_range_y -25,25 --rotation_range_z -25,25
# Noise and bias addition
python scripts/mmfit_ppda_indiv.py --noisebias --noisebias_sigma 0.2

Multi-Augmentation Assessment (Paper Section IV.C)

Run the following command to assess multiple augmentations (e.g. MM-Fit dataset):

# Run with default configuration (all augmentations enabled)
python scripts/mmfit_ppda_multi.py

Note: For other datasets, use the corresponding scripts (e.g., realdisp_ppda_indiv.py, realworld_ppda_multi.py).

📚 Citation

🙏 Acknowledgements

Parts of this repository are adapted from dl_har_public.
We thank the authors for making their code available.