🧠 Mattex Image Search Tool

A smart desktop tool for image similarity search and natural language text search, powered by SigLIP, OpenCLIP, MobileNetV4, and PyQt.

---

📦 Available Versions

🚀 v3-SigLIP2-LanceDB ⭐ (RECOMMENDED 2026)

State-of-the-art multimodal vector search engine featuring Google's SigLIP 2 and LanceDB:

• Models: SigLIP 2 SO400M (1152-dim, 384×384) & SigLIP 2 Base (768-dim, 256×256)
• Latent Space: Unified single-space vision-language encoder with native multilingual support (>100 languages)
• Database: LanceDB Embedded Vector Store (Apache Arrow zero-copy backend, ACID transactional)
• Filtering: SQL DataFusion Pre-filtering for categories and subcategories
• GUI: Modern dark theme PyQt5 interface with async thread pool, drag & drop, live progress, and cross-platform desktop integration

➡️ View README for v3-SigLIP2-LanceDB →

---

🔹 v2-Clip-Integration (Legacy)

Experimental version with text search via OpenAI CLIP ViT-B/32 + MobileNetV2.

➡️ View README for v2 →

---

🔹 v1-MobileNet (Legacy)

Classic image similarity search using TensorFlow MobileNetV2.

➡️ View README for v1 →

---

⚖️ Version Comparison

Goal	Recommended Version
State-of-the-art Multilingual Text & Visual Search	v3-SigLIP2-LanceDB (SigLIP 2 SO400M / Base)
Transactional Vector DB & SQL Filtering	v3-SigLIP2-LanceDB (LanceDB)
Legacy CLIP	v2-Clip-Integration
Legacy MobileNetV2	v1-MobileNet

---

📄 License

Released under the MIT License.
© 2026 Mattex — Feel free to use, modify, and share with credit. ✌️