📦 BlyncsySFT

Supervised Fine-Tuning for Faster R-CNN with Focal Loss and Custom Augmentations
An extensible training and evaluation framework for object detection on COCO-format datasets.

🚀 Features

• 🧠 Fine-tuning with Focal Loss for class imbalance
• 🎯 Customizable anchor boxes and backbone networks
• 🧪 Augmentation pipeline (MixUp, transforms, etc.)
• 📊 Validation pipeline with mAP evaluation (COCO)
• 🛠 CLI interface for training automation
• 🗂 Compatible with COCO-style datasets

📁 Installation

pip install blyncsysft

or

git clone https://github.com/your-username/BlyncsySFT.git
cd BlyncsySFT
pip install .

🧩 Quick Start

1. Prepare your dataset and project structure: Ensure your dataset is in COCO format. The directory structure should look like this:

   your_project/
   ├── images/
   │   ├── train/
   │   └── validation/
   ├── annotations/
   │   ├── train.json
   │   └── validation.json
   └── .env
   

2. Create a .env file: This file should contain the following variables:

   TRAINING_RUN=test01
   EPOCHS=20
   BATCH_SIZE=4
   WORKERS=2
   NUM_CLASSES=2
   BACKBONE=resnet50
   SAVE_EVERY=5
   TRAIN_IMAGE_PATH=images/train
   TRAIN_ANNOT_PATH=annotations/train.json
   VAL_IMAGE_PATH=images/validation
   VAL_ANNOT_PATH=annotations/validation.json
   

🧪 Usage

Run training directly from the command line:

python -m BlyncsySFT.cli train /path/to/your_project/ --verbose

This command:

• Valdiates your .env file
• Loads the dataset
• Builds and trains the model
• Logs the training process
• Saves the model checkpoints

Or use the Python API:

from BlyncsySFT.pipeline import run_auto_training_pipeline
from BlyncsySFT.config import load_and_validate_env

# Step 1: Load config
cfg = load_and_validate_env(env_file="/path/to/your_project/.env")

# Step 2: Define project directory
project_dir = "/path/to/your_project"

# Step 3: Run the training pipeline
run_auto_training_pipeline(project_dir, cfg, verbose=True)

📄 License

MIT License. See the LICENSE file for details.

🤝 Contributing

We welcome contributions! Please follow these steps:

1. Fork the repo and create your branch (git checkout -b feature/YourFeature).
2. Make your changes, add tests, and commit them (git commit -m 'Add some feature').
3. Submit a pull request and describe your changes.

👥 Contributors

• Sonny Jones – sonny.jones@utah.edu
• Tony Le – anthony.le@utah.edu
• Rohit Raj – rohitrraj284@gmail.com