Title:   Data-driven Inverse Kinematics using Laban Movement Analysis
Authors: Mehmet Akif Şahin, Sinan Sonlu, Uğur Güdükbay
Venue:   Computers & Graphics (Special Section on SMI 2026)

Operating system used for development and testing:
    macOS 15 (Apple Silicon, arm64). The code also runs on Ubuntu 20.04+ and on
    CPU/CUDA; only one OS is required for the replicability stamp.

How to build and reproduce:
    1. ./install.sh                       # creates ./.venv and installs deps
    2. ./.venv/bin/python reproduce.py    # regenerates the representative result

Representative result reproduced:
    The LMA descriptor-effect result of paper Figs 4-8. reproduce.py runs the
    proposed Interpolator -> Synthesizer pipeline on base motions at low vs high
    values of each LMA style descriptor (V, H, P) and reports the measured
    descriptor of the generated motion (high > low confirms the effect), plus
    exports the generated full-body joint positions to results/.

Weights: checkpoints/interpolator.pth and checkpoints/synthesizer.pth are
    included in the repository.
Dataset: obtain the LMA Effort BVH files and place the zip in data/ (see
    data/README.md); reproduce.py extracts and caches it on first run.