CoachMe

Decoding Sport Elements with a Reference-Based
Coaching Instruction Generation Model

ACL 2025

Wei-Hsin Yeh1,3, Yu-An Su1, Chih-Ning Chen1, Yi-Hsueh Lin1,2,
Calvin Ku2, Wen-Hsin Chiu2, Min-Chun Hu2, Lun-Wei Ku1,
Institute of Information Science, Academia Sinica1
National Tsing Hua University2
National Taiwan University3
{weihsinyeh168, allen0512911}@gmail.com
{yuansu, andrewman71, lwku}@iis.sinica.edu.tw
calvinku@gapp.nthu.edu.tw, whchiu@mx.nthu.edu.tw, anitahu@cs.nthu.edu.tw
Paper arXiv Code Dataset

We recommend you to read our paper in arXiv since it provides better quality of figures.

CoachMe





CoachMe generate high sport-specific instructions on Figure Skating and Boxing videos. We integrate the attention mechanism focused on the skeletal graph into the CoachMe for generating sports instruction.


The attention mechanism is visualized: Each video has 3 sets of attention focusing on different key points of the skeletal graph. Our goal is to illuminate the significance of every joint, shedding light on their individual importance. We also highlight the 3 most crucial relationships between two joints.

We select four frames from the boxing test video and arrange them in chronological order. Overlay the attention maps generated by Human Pose Perception, highlighting only the top three most important joints and the top three most significant connections between these key joints. There will be four attention maps corresponding to the four attention graphs produced by Human Pose Perception.

Each set of images is accompanied by three instructional prompts, each generated by a different model: CoachMe, LLaMA, and GPT-4o, providing their predicted guidance based on the visual content.

Additionally, include visual representations of six sport indicators: error detection, temporal awareness, body part localization, causality, methodology detection, and coordination. These six sport indicators are proposed in the CoachMe paper and are used to evaluate the relevance and applicability of the instructional prompts in the context of athletic performance.

How can an AI model coach intense sport?

AAAI-25 Educational AI Video Winner

Abstract

Motion instruction is a crucial task that helps athletes refine their technique by analyzing movements and providing corrective guidance. Although recent advances in multimodal models have improved motion understanding, generating precise and sport-specific instruction remains challenging due to the highly domain specific nature of sports and the need for in formative guidance. We propose CoachMe, a reference-based model that analyzes the differences between a learner’s motion and a reference under temporal and physical aspects. This approach enables both domain-knowledge learning and the acquisition of a coach-like thinking process that identifies movement errors effectively and provides feedback to explain how to improve. In this paper, we illustrate how CoachMe adapts well to specific sports such as skating and boxing by learning from general movements and then leveraging limited data. Experiments show that CoachMe provides high-quality instructions instead of directions merely in the tone of a coach but without critical information. CoachMe outperforms GPT-4o by 31.6% in G-Eval on figure skating and by 58.3% on boxing. Analysis further confirms that it elaborates on errors and their corresponding improvement methods in the generated instructions.

Overall Framework


Overall framework of CoachMe. CoachMe architecture comprises three modules: Concept Difference (Sec. 3.1), Human Pose Perception (Sec. 3.2), and Instruct Motion (Sec. 3.3). Instruct Motion compares the motion Tokenlearner with Tokenref to obtain the difference Tokendiff and take Tokenlearner and Tokendiff as input to the LM to generate instructions.

Distribution of Sport Indicators


By analyzing the proportion of each sport indicator present in the instructional prompts across the entire dataset (train + test), we obtain the matrix on the far left, where each value represents a proportion.

We also analyze the distribution of sport indicators predicted in the instructional prompts generated by different models—CoachMe, LLaMa, and GPT-4o—based on videos from the test dataset, and incorporate the G-eval consistency scores, which assess consistency with the ground truth. These analyses result in the three matrices on the right.

Each value in these matrices represents the total G-eval score accumulated across all instructional prompts in which the two corresponding sport indicators co-occur, normalized by the total number of prompts multiplied by the maximum possible G-eval score.