Human activity recognition (HAR) is increasingly important for healthcare, well-being, and daily monitoring applications, for which detecting alimentary activities such as eating and drinking can provide actionable insight into dietary habits and chronic disease management.
HAR systems, however, often underperform on subtle and underrepresented classes, limiting their utility in real-world dietary monitoring.
This work builds upon CABiGRU
This work builds upon CABiGRU, a convolutional architecture with Bidirectional GRU layers, multi-head attention, and residual connections, designed to capture discriminative temporal patterns from smart-watch accelerometer, gyroscope, and magnetometer data.
Improving generalization and reducing underfitting
To improve CaBiGRU's generalization and reduce underfitting in the minority class, we leverage synthetic sensor data windows using a diffusion model and adopt a two-stage training strategy: pre-training CABiGRU on synthetic data, followed by fine-tuning on the real-world data.
Results on DEO dataset
On the DEO (drinking/eating/other) dataset, the proposed pipeline achieves a balanced accuracy of 90.6%, improving over a strong supervised baseline and showing the benefits of diffusion-based synthetic pre-training for recognizing alimentary activities and representing a step forward dealing with unbalanced classes.
Enhancing robust recognition of dietary behaviors
These results suggest that combining diffusion-generated data with targeted fine-tuning enhances robust recognition of dietary behaviors, supporting more reliable deployment in healthcare and nutrition-monitoring settings.