Token Detection Model

This repository provides a framework to train a custom object detection model for recognizing various Solana tokens using Apple's CreateML.

Repository Structure

• Annotations/: Contains annotation files exported from RectLabel.
• Generated/: Output directory for the exported CreateML JSON file and trained CoreML model.
• Tokens/: Contains the download script and subdirectories for each token with their respective images.
• TokenDetector.mlproj: CreateML project file for training the object detection model.

Prerequisites

• Python 3.x: Ensure Python is installed on your system.
• icrawler: Install using pip install icrawler.
• RectLabel Pro: Available on the Mac App Store for image annotation.
• CreateML: Comes with Xcode - ensure you have the latest version installed.

Instructions

Step 1: Download Token Images

Run the download.py script inside the Tokens directory to fetch token images:

./download.py "solana token" "SOL" --max_num 100

• Search Query: Search query for the token (e.g., "Solana token")
• Folder Name: Folder name for images (e.g., "SOL")

Step 2: Annotate Images with RectLabel Pro

1. Open RectLabel Pro
2. Open images folder and annotations folder
   • Set images folder to Tokens directory
   • Set annotations folder to Annotations directory
3. Draw bounding boxes around tokens and label them
4. Export CreateML JSON file to Generated directory
   • Make sure Split to train/val/text folders is checked

Step 3: Train the Model with CreateML

1. Open TokenDetector.mlproj
2. Create a new Model Source and version it accordingly (e.g v1, v2, v3, etc.)
3. Select train for the training data
4. Press Train to start training the model

Step 4: Save the Trained Model

1. Go to Output tab after training completes
2. Click Get to save the .mlmodel file to the Generated directory