LeatherMind
A mobile-friendly leather texture recognition app with camera capture, model-backed classification, visual comparison, and scan history.
The question
Texture classification results are easier to review when the original scan, reference image, and alternative matches stay visible together.
A connected workflow.
A camera-and-gallery workflow sends images to a local Python inference backend, then presents the best match, confidence, and similar alternatives.
- 01Capture a texture
- 02Crop the region
- 03Run classification
- 04Compare the matches
In practice
- 01Camera capture, gallery upload, and an adjustable crop frame
- 02Best match and top similar results with visual verification
- 03Searchable scan history with notes and deletion controls
- 04Chinese/English and light/dark interface modes
Room to keep exploring.
The static frontend needs a backend with Python dependencies and model files for local-model inference. A prediction’s confidence is not a measured overall accuracy. The public frontend currently has missing media.
How should an interface communicate uncertainty when two textures look alike?
Project facts are grounded in the public repository. The question above is an open direction, not a reported research finding.