Skip to main content

VAME App

PyPI version

VAME App is a no-code application that lets you run the entire VAME pipeline from your browser — no scripting required.

VAME App

If you prefer working in code, the Getting Started guide covers the Python API instead. VAME App and the Python library operate on the same project format, so you can move between them freely.

Requirements​

  • Python ≥ 3.12
  • ffmpeg (needed for some video/image functions)
  • A modern browser (Chrome, Edge, Firefox, or Safari)

Installation​

pip install vame-app

Running the app​

vame-app

This starts the local server and opens the app in your browser. To stop it, return to the terminal and press Ctrl+C.

Command-line options​

vame-app [--host HOST] [--port PORT] [--data-root DIR] [--no-browser] [--dev]
FlagDefaultPurpose
--host127.0.0.1Interface to bind. Use 0.0.0.0 to expose on the LAN.
--port8641Port to listen on. Use 0 to auto-pick a free port.
--data-roothome directoryRoot the in-app file browser may traverse. Restrict this on shared servers.
--no-browseroffDon't auto-open the browser.
--devoffUse the Flask dev server instead of waitress.

The same options are available as the environment variables VAME_HOST, VAME_PORT, and VAME_DATA_ROOT.

info

Projects are always stored under ~/vame-app/projects. The --data-root option only controls which folders the in-app file browser can reach — it does not change where projects are saved.

The workflow​

VAME App walks you through the VAME pipeline step by step. Each step exposes the relevant parameters with sensible defaults, runs in the background, and reports its status so you always know where a project stands.

  1. Create a project — point the app at your pose-estimation files and set the project basics.
  2. Preprocess — clean low-confidence keypoints, remove outliers, and filter the pose data.
  3. Create the training set — choose which keypoints to include and split the data into train/test sets.
  4. Train & evaluate the model — train VAME and inspect the loss curves and reconstruction quality.
  5. Segment behavior — run motif segmentation (HMM and/or k-means) and group motifs into communities.
  6. Visualize & report — generate UMAP embeddings, motif and community videos, and summary reports.