SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports

Deeply understanding sports requires an intricate blend of fine-grained visual perception and rule-based reasoning - a challenge that pushes the limits of current multimodal models. To succeed, models must master three critical capabilities: perceiving nuanced visual details, applying abstract sport rule knowledge, and grounding that knowledge in specific visual evidence. Current sports benchmarks either cover single sports or lack the detailed reasoning chains and precise visual grounding needed to robustly evaluate these core capabilities in a multi-sport context. To address this gap, we introduce SportR, the first multi-sports large-scale benchmark designed to train and evaluate MLLMs on the fundamental reasoning required for sports intelligence. Our benchmark provides a dataset of 5,017 images and 2,101 videos. To enable granular evaluation, we structure our benchmark around a progressive hierarchy of question-answer (QA) pairs designed to probe reasoning at increasing depths - from simple infraction identification to complex penalty prediction. For the most advanced tasks requiring multi-step reasoning, such as determining penalties or explaining tactics, we provide 7,118 high-quality, human-authored Chain of Thought (CoT) annotations. In addition, our benchmark incorporates both image and video modalities and provides manual bounding box annotations to test visual grounding in the image part directly. Extensive experiments demonstrate the profound difficulty of our benchmark. State-of-the-art baseline models perform poorly on our most challenging tasks. While training on our data via Supervised Fine-Tuning and Reinforcement Learning improves these scores, they remain relatively low, highlighting a significant gap in current model capabilities. SportR presents a new challenge for the community, providing a critical resource to drive future research in multimodal sports reasoning.

Co-Pulsing FDA Radar

Target localization based on frequency diverse array (FDA) radar has lately garnered significant research interest. A linear frequency offset (FO) across FDA antennas yields a range-angle dependent beampattern that allows for joint estimation of range and direction-of-arrival (DoA). Prior works on FDA largely focus on the one-dimensional linear array to estimate only azimuth angle and range while ignoring the elevation and Doppler velocity. However, in many applications, the latter two parameters are also essential for target localization. Further, there is also an interest in radar systems that employ fewer measurements in temporal, Doppler, or spatial signal domains. We address these multiple challenges by proposing a co-prime L-Shaped FDA, wherein co-prime FOs are applied across the elements of L-shaped co-prime array and each element transmits at a non-uniform co-prime pulse repetition interval (C$^3$ or C-Cube). This co-pulsing FDA yields significantly large degrees-of-freedom (DoFs) for target localization in the range-azimuth-elevation-Doppler domain while also reducing the time-on-target and transmit spectral usage. By exploiting these DoFs, we develop C-Cube auto-pairing (CCing) algorithm, in which all the parameters are ipso facto paired during a joint estimation. We show that C-Cube FDA requires at least $2\sqrt{Q+1}-1$ antenna elements and $2\sqrt{Q+1}-1$ pulses to guarantee perfect recovery of $Q$ targets as against $Q+1$ elements and $Q+1$ pulses required by both L-shaped uniform linear array and L-shaped linear FO FDA with uniform pulsing. Numerical experiments with our CCing algorithm show great performance improvements in parameter recovery, wherein C-Cube radar achieves at least $15\%$ higher target hit-rate with shorter dwell time than its uniform counterparts.