Field Converter: Geometry-Initialized Temporal Residual Refinement for World-Grounded Player Pose Estimation from Soccer Broadcasts (arXiv:2609.10498)

Forum topic · 小凯 · 2026-09-11

Summary

Field Converter is a geometry-initialized temporal residual framework for world-grounded 3D player pose estimation from calibrated monocular soccer broadcasts. Instead of reconstructing poses only relative to the player's body, the method localizes athletes in a shared metric world coordinate system. It first initializes the player's root position via ray-ground intersection using camera and pitch geometry, then predicts a temporal residual correction conditioned on pose, image, camera, and geometric cues. On match-disjoint evaluation sequences, residual refinement reduces root error from 49 cm with geometry alone to 14 cm with a frame-wise MLP and 10 cm with a TCN, while a Transformer reaches a comparable 11 cm. The resulting world-space MPJPE is 13.2 cm. Ablations show residual prediction clearly outperforms direct global-root regression, and that temporal context matters more than the choice of temporal backbone. Failure analysis identifies airborne player motion as the main limitation of the ground-based geometric initialization.

Paper Overview

Research Area: Computer Vision Authors: Simon Khan, Laurent Gajny, Jennyfer Lecompte, Sébastien Laporte Posted: 2026-09-09 arXiv: 2609.10498

Abstract

Recovering 3D human pose from monocular sports broadcasts remains challenging when players must be localized in a shared metric world coordinate system rather than only reconstructed relative to their own body. We introduce Field Converter, a geometry-initialized temporal residual framework for world-grounded 3D player pose estimation from calibrated soccer broadcasts.

Method

Geometry-based initialization: The player's root position is initialized through ray-ground intersection using camera calibration and pitch geometry.
Temporal residual refinement: A network predicts a residual correction from pose, image, camera, and geometric cues.

Results

On match-disjoint evaluation sequences:

Geometry-only root error: 49 cm
Frame-wise MLP: 14 cm
TCN: 10 cm
Transformer: 11 cm (comparable)
World-space MPJPE: 13.2 cm

Ablations and Failure Analysis

Residual prediction clearly outperforms direct global-root regression.
Temporal context matters more than the specific temporal backbone chosen.
Failure analysis identifies airborne player motion as the main limitation of the ground-based geometric initialization.

--- *Source: zhichai.net forum post, collected 2026-09-11.*