Summary
CalTennis is a large-scale video benchmark for evaluating monocular-to-3D human pose estimation in the wild, introduced by researchers at Caltech (arXiv:2506.17579). The dataset contains footage of 40 tennis players practicing and competing, totaling over 11 million frames (about 51 hours), captured by 2 to 6 synchronized cameras at 60 Hz. It is roughly 10x larger than existing in-the-wild human motion video datasets and about 3x larger than datasets with motion-capture ground truth, and is the first large-scale benchmark providing synchronized multi-view recordings of expert athletic motion. The multi-view setup enables low-cost, markerless evaluation of monocular 3D pose algorithms, supported by a standardized capture protocol and fully automatic calibration and synchronization. Benchmarking state-of-the-art monocular 3D pose methods on CalTennis shows that 3D joint angle recovery has become fairly accurate, while models remain uncertain in depth estimation and foot contact. The authors also propose two new performance metrics—footwork and stability—and qualitatively examine body-shape inconsistency, revealing previously unexplored failure modes and concrete directions for improving pose estimation and motion analysis.
Paper Overview
Research Area: cs.CV
Authors: Ilona Demler, Xinran Xie, Blake Werner
Published: 2026-06-21
arXiv:
2506.17579Abstract
CalTennis is a large-scale video benchmark designed to evaluate monocular-to-3D pose estimation performance in the wild.
CalTennis collects footage of 40 tennis players during practice and matches — over 11 million frames totaling 51 hours — captured by 2 to 6 synchronized cameras at 60 Hz. It is more than 10x larger than existing in-the-wild human motion video datasets, and roughly 3x larger than datasets with motion-capture (MOCAP) ground-truth annotations. It is also the first large-scale benchmark to provide synchronized multi-view recordings of expert athletic motion.
The multi-view setup makes it possible to evaluate monocular 3D pose estimation algorithms in a low-cost, markerless way. The authors describe a simple, standardized capture protocol that requires no specialized equipment or expertise, complemented by fully automatic video calibration and synchronization.
Benchmarking current state-of-the-art monocular 3D pose methods on CalTennis reveals that:
- Recovery of 3D joint angles has become reasonably accurate.
- Models still struggle with depth estimation and foot-contact prediction.
The authors further propose two novel performance metrics —
footwork and
stability — and qualitatively investigate body-shape inconsistency. These metrics expose previously unexplored failure modes and point to concrete directions for improving pose estimation and motion analysis.
---
*Auto-collected on 2026-06-21*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178207995