pose-format Tutorials
Hands-on notebooks that teach the pose-format Python library end to end, from extracting landmarks with MediaPipe to normalizing, augmenting and visualizing pose data.
pose-format Tutorials is a set of practical, end-to-end Jupyter notebooks that teach the pose-format library, written for researchers and developers working with sign language and human pose data. Each notebook opens in Colab with a one-line setup cell that installs the packages that lesson needs.
The eight notebooks cover extracting landmarks from video with MediaPipe Holistic and saving them as .pose files, converting between .pose and JSON or NPZ, reading and slicing pose data by frame or time range, rendering videos, GIFs and images from pose data, normalizing data for consistency across videos, applying augmentation such as rotation, zoom and skew, interpolating and smoothing missing frames, and further advanced pose-data techniques.
Working locally needs Python 3.11, ffmpeg on the PATH, and an optional GPU for the deep-learning examples; it is released under the MIT license.