How Good Was My Shot? Quantifying Player Skill Level in Table Tennis (arXiv 2603.25736)

Forum topic · 小凯 · 2026-03-28

Summary

This post summarizes the paper "How good was my shot? Quantifying Player Skill Level in Table Tennis" (arXiv:2603.25736) by Akihiro Kubota, Tomoya Hasegawa, Ryo Kawahara, and Ko Nishino in the field of computer vision. The authors tackle the challenge of quantifying latent skill from observed behavior by studying dyadic sports, focusing on table tennis. Their key idea is to learn generative models of each player's tactical racket strokes and jointly embed them in a common latent space that encodes individual characteristics, including skill level. The models are trained on a large-scale dataset of 3D-reconstructed professional matches and conditioned on comprehensive game context such as player positioning and opponent behavior. Probing this learned player space reveals distinct playing styles and attributes that jointly represent skill level. Training a simple relative ranking network on these embeddings enables both relative and absolute skill prediction. The results show that the learned player space effectively quantifies skill, laying groundwork for automated skill assessment in complex interactive behaviors.

Paper Overview

Field: Computer Vision (CV) Authors: Akihiro Kubota, Tomoya Hasegawa, Ryo Kawahara, Ko Nishino Published: 2026-03-26 arXiv: 2603.25736

Summary (Chinese forum translation)

Gauging an individual's skill level is crucial, as it inherently shapes their behavior. Quantifying skill, however, is challenging because it is latent to the observed actions. To explore skill understanding in human behavior, the authors focus on dyadic sports — specifically table tennis — where skill manifests not just in complex movements, but in the subtle nuances of execution conditioned on game context.

The key idea is to learn a generative model of each player's tactical racket strokes and jointly embed them in a common latent space that encodes individual characteristics, including those pertaining to skill levels. By training these player models on a large-scale dataset of 3D-reconstructed professional matches and conditioning them on comprehensive game context — including player positioning and opponent behavior — the models capture individual tactical characteristics within their latent space.

Probing this learned player space shows that it reflects distinct playing styles and attributes that jointly represent skill level. By training a simple relative ranking network on these embeddings, the authors demonstrate that both relative and absolute skill prediction can be achieved.

Original Abstract (excerpt)

> Gauging an individual's skill level is crucial, as it inherently shapes their behavior. Quantifying skill, however, is challenging because it is latent to the observed actions. To explore skill understanding in human behavior, we focus on dyadic sports -- specifically table tennis -- where skill manifests not just in complex movements, but in the subtle nuances of execution conditioned on game context. Our key idea is to learn a generative model of each player's tactical racket strokes and jointly embed them in a common latent space that encodes individual characteristics, including those pertaining to skill levels. By training these player models on a large-scale dataset of 3D-reconstructed professional matches and conditioning them on comprehensive game context -- including player positi...

*Auto-collected on 2026-03-28*