PhotoLense

Desktop photo manager with on-device AI classification

Browse your filesystem, view photos with EXIF metadata, and classify images into categories using on-device AI models — no cloud, no API keys, fully private.

---

Screenshot placeholder — add your screenshot to docs/screenshot.png

---

Highlights

• AI-Powered Classification — Run ConvNeXt V2 or MobileNetV3 models locally via ONNX Runtime with GPU acceleration (DirectML / CoreML / CUDA)
• Find Similar Photos — Cosine similarity search on model embeddings powered by sqlite-vec
• Group by Color — Organize photos by dominant color using CIELAB color space with Fixed Palette or K-Means clustering
• EXIF Metadata Panel — Camera make/model, exposure, aperture, ISO, GPS, and more
• Blazing Fast Browsing — Lazy-loaded thumbnails via IntersectionObserver, handles thousands of images without freezing
• 100% Offline — Everything runs on your machine. No cloud. No telemetry.

Features

Photo Browsing

• Grid view and List view toggle
• Lazy-loaded thumbnails — only generates thumbnails for photos visible in the viewport
• Sort by name, size, or date (ascending/descending)
• Real-time filename and tag search
• File watcher for auto-refresh when files change externally (debounced ~300ms)
• Tag filtering with multi-select dropdown

AI Classification

• Models: ConvNeXt V2 Base/Large (ImageNet-22K, 384x384) and MobileNetV3 Large (ImageNet-1K, 224x224)
• Runtime: ort (ONNX Runtime) with GPU acceleration
• Automatic model download on first use
• Top-K predictions with confidence scores
• Optional auto-organize: move or copy images into subfolders by predicted class
• Embeddings stored in SQLite for similarity search

Find Similar

• Select a photo and find visually similar images using cosine similarity on model embeddings
• Non-blocking background indexing with progressive results
• Adjustable similarity threshold via toolbar slider

Group by Color

• Organize photos by dominant color in CIELAB color space
• Fixed Palette or K-Means Clustering (via kentro) with configurable cluster count

EXIF Metadata

• Toggle-able info panel with camera details
• Supports: camera make/model, date taken, exposure time, aperture, ISO, focal length, dimensions, orientation, GPS coordinates, flash, white balance, software
• Histogram display

Architecture

Backend (Rust): Tauri commands expose filesystem operations, thumbnail generation, EXIF parsing, and ML inference. The classifier uses ONNX Runtime (via ort) with ImageNet normalization (NCHW format, softmax post-processing). Embeddings are stored in sqlite-vec for cosine similarity search. Heavy tasks (thumbnails, color extraction) are parallelized with rayon.

Frontend (React + TypeScript): State managed via useReducer + React Context. No external UI libraries. Thumbnails load lazily per-item using IntersectionObserver with an in-memory cache.

Tech Stack

Layer	Technology
Framework	Tauri 2
Frontend	React 18, TypeScript 5.6, Vite 6
Backend	Rust (2021 edition)
ML Runtime	ort (ONNX Runtime — DirectML, CoreML, CUDA)

Getting Started

Prerequisites

• Node.js v18+
• Rust (stable)
• Tauri CLI prerequisites for your platform

Install & Run

# Clone the repository
git clone https://github.com/srad/PhotoLense.git
cd PhotoLense

# Install frontend dependencies
npm install

# Run in development mode
npm run tauri dev

# Build for production
npm run tauri build

How Classification Works

1. Open a folder and click Classify in the toolbar
2. Select a model (MobileNetV3 Large, ConvNeXt Base, or ConvNeXt Large) and configure options (top-K, organize, GPU)
3. Click Start Classification
4. The app downloads the ONNX model (first time only), loads it via ONNX Runtime, then for each image:
   • Resizes and center-crops to the model's input size (224x224 or 384x384)
   • Normalizes with ImageNet mean/std (NCHW tensor)
   • Runs inference and applies softmax for top-K class predictions
   • Stores L2-normalized logit embeddings in sqlite-vec for similarity search
5. If "Organize" is enabled, images are moved or copied into subfolders named by their top predicted class

Cross-Platform

PhotoLense runs on Windows, macOS, and Linux. ONNX Runtime binaries are downloaded automatically during build. Drive listing uses the Windows API on Windows and falls back to / on Unix systems.

Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/my-feature)
3. Commit your changes (git commit -m 'Add my feature')
4. Push to the branch (git push origin feature/my-feature)
5. Open a Pull Request

License

PolyForm Noncommercial License 1.0.0 means it is free for non-commecial purposes.