DUET-AFDL: A Computational Resource for Anhui Flower-Drum Lantern Dance

DUET-AFDL is the first comprehensive computational resource for Anhui Flower-Drum Lantern (AFDL) dance, a representative Chinese intangible cultural heritage dance. This repository implements the Digital Unification of Embodied Training (DUET) framework. Note: As this repository represents the first open-source computational resource for AFDL, we are continually refining the English translations. We welcome feedback and suggestions for improving the accuracy of cross-cultural terminology. File Structure The DUET-AFDL repository implements the aforementioned architecture, and its file organization is as follows: duet-afdl-v1/ ├── metadata/ │ ├── single_actions_metadata.csv │ ├── dance_phrases_metadata.csv │ ├── bilingual_terminology_list.csv │ ├── contributors.csv │ └── music_01_9x8beats_24s.mp3 ├── single_action/ │ ├── 1001/ │ │ ├── [ID]_demo.mp4 │ │ ├── [ID]_annotation.csv │ │ ├── [ID]_teacher_001_imu.pkl │ │ ├── [ID]_teacher_002_imu.pkl │ │ ├── [ID]_student_001_imu.pkl │ │ ├── [ID]_student_001_annotation.csv │ │ └── ... (additional student files) │ ├── ... (Other fundamental movements) ├── dance_phrase/ │ ├── 20250630001/ │ │ ├── [ID]_demo.mp4 │ │ ├── [ID]_annotation.csv │ │ └── [ID]_choreography_notes.md │ ├── ... (Other dance phrases) ├── figures/ └── README.md DUET-AFDL currently encompasses the following comprehensive data. Category Count Description Fundamental Movements 51 Fundamental movement units of Anhui Flower-Drum Lantern dance across 4 difficulty levels Dance Phrases 22 Choreographed sequences (64 beats each) Expert Videos 73 High-quality demonstration videos Expert IMU Samples 102 2 IMU recordings per action Student IMU Samples 204 4 IMU recordings per action Expert Annotations 277 Beat-level expert labels and error markings Bilingual Terms 55 Standardized Chinese-English terminology Data Format and Specifications The repository follows strict naming conventions to ensure consistency and support automated processing. When contributing new data to this repository, please adhere to the naming conventions described in this section. To facilitate querying and data processing, we assign unique IDs to fundamental movements, dance phrases, students, experts, and so on. Object ID Fundamental Movement Four-digit codes ranging from 1001 to 4023, where the first digit indicates difficulty level (1-4) Dance Phrases Date-based codes in the format YYYYMMDDNNN (e.g., 20250630001) Experts Consists of four characters, with the first character being T (teacher) followed by three digits, such as T001 Students Consists of four characters, with the first character being S followed by three digits, such as S001 Metadata Based on past teaching experience, we have defined 51 fundamental movements of AFDL, which represent the core and most basic elements of AFDL. The metadata/single_actions_metadata.csv file contains the canonical definition for each fundamental movement: Header Description id The fundamental movement ID name_zh Chinese name of the movement name English name of the movement contributor_id The ID of the contributor who designed this fundamental movement level Difficulty (1=Basic, 2=Intermediate, 3=Challenging, 4=Difficult) category_zh Category name in Chinese category Category name in English, for retrieval queries, teaching category classification, etc. description_zh Chinese description of the movement, include the design background, characteristics, and difficulties of this action description The English translation of the Chinese description of the movement audio Accompaniment music filename The fundamental movements currently created constitute the core of AFDL, primarily targeted at beginners. For more advanced practitioners, additional entries may be expanded in the future. On the basis of these fundamental movements, we further choreographed 22 dance phrases. The metadata/dance_phrases_metadata.csv file contains the canonical definition for each expert-created dance phrase: Header Description id The dance phrase ID name_zh Chinese name of the dance phrase name English name of the dance phrase contributor_id The ID of the contributor who designed this dance phrase level Difficulty (1=Basic, 2=Intermediate, 3=Challenging, 4=Difficult) beats The total number of beats in the dance phrase constituent_action_id The fundamental movement sequence composing this dance phrase (separated by ;, e.g., 1001;1002;1005;1003) constituent_action_zh The Chinese version of the aforementioned sequence description_zh Chinese description of the dance phrase, include the design background, characteristics, and difficulties of this action description The English translation of the Chinese description of the dance phrase audio Accompaniment music filename For each dance phase, experts have written a Markdown-formatted choreography note, recording as much detail as possible about the relevant cultural background, purpose, and teaching notes. The documentation file is named [ID]_choreography_notes.md. The dance phrases also supports expansion, and teachers can create entries as needed according to the actual situation. When adding a new entry, please ensure that all the above-defined headers are included in the metadata file. Sensor Data For the acquisition of motion data, we utilize the Bluetooth inertial measurement unit (IMU) WT9011DCL-BT50 from WitMotion Shenzhen Co., Ltd. The sensor is mounted at one end of the fan to track its motion, which also captures the dancer's body movements. The sensor provides the following physical quantities at time $t$: $$\vec{s}(t) = [a_x(t), a_y(t), a_z(t), \omega_x(t), \omega_y(t), \omega_z(t), \theta_x(t), \theta_y(t), \theta_z(t)]^\mathsf{T}$$ where the values of $(\theta_x, \theta_y, \theta_z)$ are obtained by integrating the angular velocities. The units of each physical quantity are as follows: Physical Quantity Unit $a_x, a_y, a_z$ Multiples of $g$ $\omega_x, \omega_y, \omega_z$ rad/s $\theta_x, \theta_y, \theta_z$ rad We record the fan movement data at a sampling rate of 20Hz. After each data collection, we retain only the valid portion corresponding to the dancing section synchronized with music, and segment the data into units of four beats. For each fundamental movement, we have collected two samples from experts (teachers) and four samples from students. The data files are named [id]_[role]_[num]_imu.pkl, such as 1001_teacher_001_imu.pkl. To help dancers more clearly anchor the starting time, the accompaniment music metadata/music_01_9x8beats_24s.mp3 contains nine 8-beat measures, with the first 8-beat measure serving as a preparation period during which no actions are scheduled. When contributing new IMU data, please ensure that files adhere to the structure outlined above and, where possible, maintain consistent sampling rates and coordinate systems with existing data. Annotation Files We utilize annotation files precisely synchronized to the musical beat to record finer details. For each designed action, including fundamental movements and dance phrases, we provide a corresponding annotation file named [id]_annotation.csv: Header Description segment_index The range is 1–16, indicating the position of the corresponding 4-beat segment. For example, segment_index=2 denotes the second 4-beat position. expert_note Expert annotations for the specified 4-beat action, such as common pitfalls. overall_comment Expert commentary on the overall structure of the dance segment expert_id The ID of the expert who authored this annotation file For each student's IMU data, we also provide an annotation file written by experts, which precisely indicates the locations of errors in the student's movements. The file is named [ID]_student_[num]_annotation.csv. Header Description segment_index The position of the corresponding 4-beat segment label The value is 0 or 1, where 1 indicates an error in the 4-beat action and 0 indicates no error. expert_note Expert annotations for the student's 4-beat action overall_comment Expert commentary on the overall student performance contributor_id The ID of the contributor who contributed this IMU data sample Contributing If you use DUET-AFDL in your research, please cite: @data{duet_afdl_2025, author = {Yuquan Chen, Rui Niu, Shichun Liu}, title = {DUET-AFDL: A Computational Resource for Anhui Flower-Drum Lantern Dance}, year = {2025}, version = {v1}, publisher = {Zenodo}, doi = {10.5281/zenodo.17964799} } We welcome contributions to expand and improve the DUET-AFDL repository. Areas for contribution include: Data Expansion: Adding more teacher/student samples following existing naming conventions Action Coverage: Extending the atomic movement vocabulary with new entries in metadata/single_actions_metadata.csv Sensor Modalities: Integrating additional sensing technologies while maintaining compatibility with the DUET framework Annotations: Providing expert annotations for unlabeled data using the established annotation schema Documentation: Translating documentation to other languages while preserving bilingual structure Tools: Developing analysis and visualization tools Please contact us (Email: [e-mail]) before making significant contributions to ensure alignment with the DUET framework standards and maintain consistency with existing data structures. License This project is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). For questions about licensing or commercial use, please contact us through the channels provided in the Contact section.