Search papers, labs, and topics across Lattice.
This paper details a system developed for the SoccerNet 2026 Player-Centric Ball-Action Spotting Challenge, focusing on identifying player actions in soccer broadcasts across eight categories. The authors enhance the existing FOOTPASS baselines through four key extensions, including gradient checkpointing for efficient fine-tuning, integration of GNN logits with visual features, and a novel post-processing pipeline to improve prediction accuracy. Their approach results in a Macro F1 score of 0.548 on the test set, demonstrating significant advancements in action recognition within soccer analytics.
Achieving a Macro F1 score of 0.548, this system effectively tackles the challenge of predicting player actions in soccer broadcasts, overcoming significant data imbalance.
We describe our system for the SoccerNet 2026 Player-Centric Ball-Action Spotting Challenge, which requires predicting who performs which action and when, across eight classes in broadcast soccer. Building on the three FOOTPASS baselines [1] (TAAD, TAAD+GNN, and TAAD+DST), we contribute four extensions: (1) gradient check pointing to enable full-backbone fine-tuning on a single GPU; (2) fusion of GNN logits into the DST encoder, combining graph-based tactical context with per-player visual features; (3) square-root frequency class weighting to address the 213:1 pass-to-tackle imbalance in the training data; and (4) a post processing pipeline comprising per-class logit gating, temporal frame refinement, jersey re-assignment, and a two-model ensemble. Our system achieves 0.548 Macro F1 on the test set and 0.446 on the challenge set (server evaluation).
