This repository contains a digital-history project on ballet technique at the Prix de Lausanne.
Contains cleaned CSV files used for the arabesque analysis. These files include processed metadata and extracted movement measurements, such as arabesque angle, torso tilt, split angle, pelvic tilt, support-leg alignment, and shoulder-hip torsion.
Main files:
arabesque_analysis_ready_updated.csvarabesque_cleaned_for_analysis.csv
Contains yearly PDFs and extracted candidate information for the Prix de Lausanne. These files are used to study the number of candidates, selected candidates, and competition participation patterns over time.
Stores outputs from the video and pose-analysis pipeline. It includes per-edition arabesque metric CSV files, extracted key frames, skeleton JSON files, and intermediate results generated from performance videos.
Contains reusable Python scripts for the movement-analysis pipeline:
extract_arabesque.py: extracts arabesque pose metrics from videos.extract_additional_features.py: computes additional movement or pose features.extract_frame_row.py: extracts frame-level information.merge_metadata.py: merges movement results with metadata.
Contains generated figures and animations for the final analysis, including decade comparisons, PCA plots, trend plots, distribution plots, and male/female arabesque evolution GIFs.
Contains helper scripts used during development, such as notebook search and pipeline-update utilities. These are not the main analysis files.
A lightweight local compatibility folder for COCO-style pose-evaluation imports used by the pose-estimation pipeline.
Local environment/editor files. They are not part of the core research analysis.
Documents the metadata-cleaning process and general competition analysis. It explains manual corrections, gender labeling, missing years, classical/contemporary variation filtering, and the construction of analysis datasets.
It also analyzes finalists by country, gender, school, year, and awards.
Runs the arabesque extraction pipeline module by module. This notebook is mainly used to process videos and generate movement metrics.
Analyzes the extracted arabesque measurements. It includes feature correlations, temporal trends, standardization/diversity analysis, generational changes, motion extremes, and 3D skeleton visualization.
Compares arabesque geometry across country/tradition groups before 2011. It groups dancers into traditions such as French, Russian/Vaganova, British/RAD, Balanchine/American, and Others, then compares six geometry metrics.
Analyzes school/style differences from 2012 to 2026. It compares school-style groups, decade patterns, geometry profiles, and whether time-period effects remain after controlling for training style.
The project uses a Conda environment with Python 3.10 and packages for data analysis, visualization, video processing, and pose estimation.
conda env create -f environment.yml
conda activate ballet_envImportant packages include pandas, numpy, matplotlib, opencv-python, moviepy, mmpose, mmdet, and scikit-learn.