First public release. Install from PyPI:
pip install hotgaze
Highlights
- Attention heatmaps for UI screenshots —
hotgaze run screenshot.png -o overlay.png - Numeric region scores (the differentiator):
hotgaze scoreemits canonical, deterministic JSON — attention share, peak, rank per region - A/B compare:
hotgaze compare a.png b.png— per-region deltas, 3×3 spatial grid, focal-point movement - Two backends:
fast(zero downloads, offline, default) anddeep(UNISAL,pip install hotgaze[deep]) - Faces layer: YuNet face detection via
--layers faces - Local-only: no cloud, no API keys, no telemetry
Model weights are hosted separately under the weights-v1 release and download automatically on first use (SHA-256 verified).
Full changelog: CHANGELOG.md