Radar Pose
The project
Working with a four-person team, I trained models to predict 17 body joints from radar data.
Preparing the data and models
The work included MMVR preprocessing with Gaussian heatmap targets, a custom CNN, and a modified ResNet18.
Results and scope
The project reached 0.931 PCK@5% on a 200-sample subset. This result describes that subset, rather than a full-dataset evaluation.