Human Pose Estimation in Trampoline Gymnastics: Improving Performance Using a New Synthetic Dataset
This addresses a domain-specific problem for trampoline gymnastics by improving pose estimation accuracy in extreme conditions, but it is incremental as it builds on existing models and datasets.
The paper tackled the problem of human pose estimation in trampoline gymnastics, where extreme poses and viewpoints cause state-of-the-art models to under-perform, by fine-tuning a ViTPose model on a synthetic dataset generated from motion capture data, resulting in state-of-the-art 2D accuracy and a 19.6% reduction in 3D MPJPE (12.5 mm improvement).
Trampoline gymnastics involves extreme human poses and uncommon viewpoints, on which state-of-the art pose estimation models tend to under-perform. We demonstrate that this problem can be addressed by fine-tuning a pose estimation model on a dataset of synthetic trampoline poses (STP). STP is generated from motion capture recordings of trampoline routines. We develop a pipeline to fit noisy motion capture data to a parametric human model, then generate multiview realistic images. We use this data to fine-tune a ViTPose model, and test it on real multi-view trampoline images. The resulting model exhibits accuracy improvements in 2D which translates to improved 3D triangulation. In 2D, we obtain state-of-the-art results on such challenging data, bridging the performance gap between common and extreme poses. In 3D, we reduce the MPJPE by 12.5 mm with our best model, which represents an improvement of 19.6% compared to the pretrained ViTPose model.