Combat-Monitor: Dual-Stream Attention Framework for Athlete Wellbeing

# Combat Sports Athlete Monitoring with Dual-Stream Attention and Temporal Calibration ## OverviewThis repository implements a novel computational framework for monitoring athlete wellbeing in combat sports. The system integrates **Dual-Stream Attention Mechanism for Combat Dynamics (DS-AMCD)** and **Interaction-Adaptive Temporal Calibration (IATC)** to achieve high-precision action detection and interaction intensity estimation under real-world conditions (e.g., occlusions, rapid movement, and lighting variations). Key contributions:- **Cross-modal fusion** of skeletal pose dynamics and contextual visual features.- **Adaptive temporal calibration** using domain knowledge like striking rhythms and guard recovery patterns.- Robust performance across diverse combat sports (boxing, MMA, kickboxing). --- ## Features- High-resolution object detection tailored to combat sports.- Multimodal inputs: **RGB frames** + **pose estimations** (joint coordinates, velocities, accelerations).- Attention-based feature fusion and hierarchical temporal reasoning.- Real-time athlete monitoring with adaptive injury prevention insights. --- ## MethodologyThe framework consists of:1. **Dual-Stream Encoder**: Parallel branches for pose and visual streams, fused via cross-modal attention.2. **Hierarchical Temporal Reasoning**: Captures both immediate and long-range behavioral trends.3. **Interaction-Adaptive Temporal Calibration**: Dynamically refines temporal predictions using domain-specific priors. Detailed architecture diagrams are provided in the paper (see Figures 1–3). --- ## Experimental ResultsEvaluations on multiple datasets (LibriSpeech, TIMIT, Common Voice, TED-LIUM) demonstrate:- Superior accuracy and F1 scores compared to YOLOv5, RetinaNet, DETR, and others.- Enhanced robustness in noisy and diverse environments.- Ablation studies confirm the synergy of multimodal encoding, cross-modal attention, and adaptive calibration. --- ## Getting Started ### Prerequisites- Python 3.8+- PyTorch >= 1.10- HuggingFace Transformers- CUDA-enabled GPU (e.g., NVIDIA A100 recommended) ### Installation```bashgit clone https://github.com/your-username/combat-sports-monitoring.gitcd combat-sports-monitoringpip install -r requirements.txt