Colorizer

Colorizer is a web service that maps natural language text to colors using sentence embeddings. It leverages a pre-trained ONNX model (distiluse-base-multilingual-cased-v2) to generate embeddings, compares them to a set of reference embeddings, and returns the closest color as JSON.

This project demonstrates how to combine ONNX models, Rust, and Actix Web for a lightweight AI-powered API.

---

Features

• Input any text in multiple languages (thanks to distiluse-base-multilingual-cased-v2)
• Returns the closest reference color as RGB (r, g, b)
• Pre-computed reference embeddings for fast lookup
• Rate-limited API with actix-governor
• Serves static frontend files under /static

---

Installation

1. Clone the repository

git clone https://github.com/8ria/Colorizer.git
cd Colorizer

2. Install Rust

Make sure you have Rust 1.70+ installed:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

3. Download models

Note: The models/ folder is not included in this repository due to size and licensing. You must download the following files and place them in models/:

• tokenizer.json — tokenizer for distiluse-base-multilingual-cased-v2
• model.onnx — ONNX model converted from distiluse-base-multilingual-cased-v2

---

Usage

1. Generate reference embeddings

You can generate embeddings for predefined words (or customize your own):

cargo run --bin generate_ref_embeddings

This will create custom/ref_embeddings.json.

2. Run the server

cargo run

The server will start at http://localhost:8090/.

---

API

POST /color

Request:

{
  "text": "sun"
}

Response:

{
  "r": 255,
  "g": 255,
  "b": 0
}

---

GET /

Serves static/index.html if available. Useful for a simple frontend.

---

Project Structure

Colorizer/
├─ src/
│  ├─ main.rs               # Actix server
│  ├─ bin/
│  │  └─ generate_ref_embeddings.rs  # Embedding generator
├─ models/                  # ONNX model + tokenizer
├─ custom/                  # Generated reference embeddings
├─ static/                  # Optional static frontend
├─ Cargo.toml
└─ README.md

---

Dependencies

• actix-web — web framework
• actix-files — static file serving
• actix-governor — rate limiting
• ort — ONNX Runtime for Rust
• ndarray — numerical arrays
• serde + serde_json — JSON serialization
• tokenizers — HuggingFace tokenizers

---

Notes

• The system uses cosine similarity to match input embeddings to reference colors.
• You can easily extend custom/ref_embeddings.json with more words/colors.
• For production deployment, consider HTTPS, caching, and scaling options.

---

License

MIT License © AndriaK