Research
TensorFlow Hidden Composition
A TensorFlow.js research project for reading photographs
through abstraction, alignment, and gradual reveal.
The launch build is now live as a TensorFlow experiment
scaffold.
What This Project Is
TensorFlow Hidden Composition is a new browser-based experiment in which a photograph is approached as structure before it is recognized as image. Each work begins as a reduced composition of color, line, and target geometry.
What TensorFlow Does Here
- Tracks body, hand, and face geometry in the browser
- Supports live alignment scoring and reveal thresholds
- Creates a practical base for future abstraction-specific models
Current Launch Build
The current launch route is wired to the existing TensorFlow experiment shell so the project can be explored, shared, and iterated as the full Hidden Composition interaction is built out.