About TriPix Viewer
TriPix Viewer is the public gallery of the TriPix photo platform. It shows photos that contributors have marked as public, organised by category and place.
The archive view is for signed-in users and includes private photos. The viewer is open to everyone and only displays public content.
Features
- Browse public photos by category and location
- Full-screen view of individual images
- Camera app for contributors to capture and upload new photos
Contact
For legal information and operator details, see Impressum.
Technical
TriPix is a small custom photo platform: a mobile camera app, a local manager on the contributor’s PC, and this public viewer on tripix.club. Photos are stored in a MySQL database; originals and thumbnails live on the webspace under each user’s hash folder.
Pipeline
From capture to the gallery, each photo passes through these stages:
- Capture & upload — The camera app sends the image with GPS coordinates (from EXIF or the phone browser). Uploads can be marked private or public.
- Download for processing — The local manager fetches low-resolution copies from tripix.club for offline work.
- Hashes & duplicates — Perceptual hashing (dHash) and sharpness scores are computed; exact duplicates are removed automatically.
- Cleanup — Similar shots are clustered; an AI pass rejects blurry, accidental, or empty frames.
- Location — GPS coordinates are reverse-geocoded to a place name (city, region, country) for filtering in the archive and viewer.
- Classification — An AI assigns one or more categories per photo. The contributor can add custom category names and re-run classification.
- Review — A drag-and-drop board lets you move photos between categories or assign several categories to one image.
- Publish — Locked classifications go live in the web archive; public photos appear here in TriPix Viewer.
AI & external services
- Qwen3-VL 8B (via Ollama, run locally on the contributor’s PC) — vision model used for cleanup (reject bad shots) and for multi-category classification.
- Nominatim / OpenStreetMap — reverse geocoding from GPS; place names in the viewer come from coordinates, not from the vision model.
No cloud AI API is called during normal processing: Qwen runs on the local machine through Ollama. Only geocoding uses an external web service (OpenStreetMap).