EpisodicLatentSpectral

EpisodicLatentSpectral Skeleton Sequence Representation Learning and Performance Quality Evaluation for Dance Movement Recognition Overview EpisodicLatentSpectral is a deep learning framework for skeleton-based dance movement recognition and quantitative performance quality assessment. Unlike conventional action recognition methods that only classify movement categories, this framework jointly learns motion representations and evaluates dance performance quality. It integrates episodic memory retrieval, latent dynamics modeling, and spectral consistency learning into a unified architecture, enabling robust temporal representation learning and interpretable quality prediction. Highlights - End-to-end framework for skeleton-based dance analysis- Episodic memory retrieval for long-range temporal dependency modeling- Latent dynamics simulation for motion representation learning- Spectral consistency modeling using frequency-domain constraints- Joint optimization for recognition and quality evaluation- Applicable to both action recognition and dance performance assessment Framework The proposed framework consists of three major components: 1. Episodic Memory Retriever Retrieves historical motion patterns from learnable memory slots, providing contextual information for each skeleton sequence. 2. Latent Dynamics Simulator Encodes skeleton sequences into a latent space and models temporal evolution for robust motion representation. 3. Spectral Consistency Mapper Preserves frequency-domain characteristics of motion sequences through FFT-based spectral constraints, improving temporal consistency and robustness. Together, these modules generate discriminative representations for both dance recognition and performance quality prediction. Features - Skeleton sequence representation learning- Dance movement recognition- Performance quality evaluation- Temporal dependency modeling- Latent motion dynamics- Spectral consistency learning- Multi-objective optimization Project Structure .├── configs/├── datasets/├── models/│ ├── episodic_memory.py│ ├── latent_dynamics.py│ ├── spectral_mapper.py│ └── model.py├── losses/├── scripts/├── train.py├── evaluate.py├── requirements.txt└── README.md Installation Clone the repository: git clone https://github.com/<username>/EpisodicLatentSpectral.gitcd EpisodicLatentSpectral Create a virtual environment: python -m venv .venvsource .venv/bin/activate Install dependencies: pip install -r requirements.txt Dataset The framework is designed for skeleton-based dance datasets represented by 2D or 3D joint coordinates. Example data format: sample/├── skeleton.npy├── label.txt└── quality.txt where - skeleton.npy: Skeleton sequence `(T × J × D)`- label.txt: Dance category- quality.txt: Performance quality score Training python train.py --config configs/train.yaml Evaluation python evaluate.py \ --config configs/eval.yaml \ --checkpoint checkpoints/best_model.pt Evaluation Metrics Recognition - Accuracy- Precision- Recall- F1-score- AUC Quality Assessment - Mean Absolute Error (MAE)- Spearman Correlation- Intraclass Correlation Coefficient (ICC)- Pairwise Ranking Accuracy Roadmap - [ ] Release complete source code- [ ] Release training scripts- [ ] Release pretrained models- [ ] Release dataset preprocessing tools- [ ] Add visualization utilities- [ ] Add inference examples Citation If you find this project useful, please cite: @article{qiu2025episodiclatentspectral, title={Skeleton Sequence Representation Learning and Performance Quality Quantitative Evaluation for Dance Movement Recognition}, author={Xiaoli Qiu}, journal={TBD}, year={2025}}