# Can Event Detectors Leave the Broadcast Camera? Evidence from Tennis

DevFeed: [Can Event Detectors Leave the Broadcast Camera? Evidence from Tennis](<https://devfeed.tech/articles/can-event-detectors-leave-the-broadcast-camera-evidence-from-tennis-108010.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.38042>)

Author: David Soeiro-Vuong, Andre-Louis Rochet

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Computer vision](<https://devfeed.tech/topics/computer-vision.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>)

Tags: [annotations](<https://devfeed.tech/tags/annotations.md>), [arbitration](<https://devfeed.tech/tags/arbitration.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [bounce](<https://devfeed.tech/tags/bounce.md>), [broadcast](<https://devfeed.tech/tags/broadcast.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [court](<https://devfeed.tech/tags/court.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [e2e](<https://devfeed.tech/tags/e2e.md>), [hit](<https://devfeed.tech/tags/hit.md>), [phone](<https://devfeed.tech/tags/phone.md>), [release](<https://devfeed.tech/tags/release.md>), [split](<https://devfeed.tech/tags/split.md>), [timing](<https://devfeed.tech/tags/timing.md>), [video](<https://devfeed.tech/tags/video.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

## AI overview

The benchmark suggests tennis event detectors transfer poorly from broadcast footage to other viewpoints, though approaches differ. It contains 345 frame-level bounce and hit annotations across eight broadcast, elevated, phone and court-level clips, with frozen splits and predictions. Supervised spotters using broadcast weights perform near perfectly on broadcast footage but poorly on phone and court-level video (E2E-Spot pooled bounce F1: 0.130). RallyArb, which combines ball tracking and player poses without fitting to event labels, exceeds 0.8 bounce F1 across three phone and court-level clips, but two were used for tuning; it finds 8 of 11 bounces on the held-out clip and misses human hit timing. Zero-shot vision-language models detect events across viewpoints but miss contact frames by about 0.1 seconds. Leave-one-clip-out fine-tuning reaches about 0.8 bounce F1 on two clips but only 0.2 on a grazing phone angle.

## Source excerpt

arXiv:2610.38042v1 Announce Type: cross Abstract: Ball bounces and racket hits are the events from which tennis statistics are derived, and on monocular video the two are easy to confuse. We release a benchmark of 345 bounce and hit annotations at frame level on eight clips that mix broadcast, fixed elevated, phone and court-level viewpoints, with a frozen development and test split, a confusion-aware matcher, clip-level bootstrap intervals and a blind second annotation of three clips. On it we score supervised precise spotters with published broadcast weights, two zero-shot vision-language models, and RallyArb, an arbitration layer over a ball track and player poses that fits no parameter on event labels. The supervised spotters score near-perfectly on broadcast footage and almost nothing on phone and court-level footage, where the pooled bounce F1 of E2E-Spot is 0.130. RallyArb keeps a bounce F1 above 0.8 on the three phone and court-level clips, two of which were used to tune it, and finds 8 of 11 bounces on the held-out one, but it places hits too far from the contact frame to match human timing. The vision-language models detect events on every viewpoint and miss the contact frame by about 0.1 s. Fine-tuned leave-one-clip-out from its broadcast weights, the spotter reaches about 0.8 bounce F1 on two phone and court-level clips and stays at 0.2 on a grazing phone angle. The annotations, the evaluator and the frozen predictions of every scored row are released.