👗 Fashion Parser 🧥

Fashion segmentation based on Mask R-CNN (ResNet-50 FPN) using the imaterialist-fashion-2019-FGVC6 dataset.

---

🛠️ Technology Stack

Category	Technology
Framework	PyTorch 2.1+, Torchvision
Backbone	ResNet-50 Feature Pyramid Network (FPN)
Package Management	uv
Serving	NVIDIA Triton Inference Server
Configuration	python-dotenv
Visualization	Matplotlib, OpenCV

---

🏁 Quick Start

1️⃣ Installation

Ensure you have uv installed, then run:

uv sync

2️⃣ Configuration

Copy the template and set your dataset paths:

cp .env.example .env

3️⃣ Start Training

uv run src/fashion_parser/scripts/train_torch.py

---

🚢 Production Deployment (Triton)

Export to TorchScript

Prepare your trained model for production:

uv run src/fashion_parser/scripts/export_triton.py --weights fashion_mrcnn_pytorch.pth

Launch Server

Deploy the model repository using Docker:

docker run --gpus=all --rm -p 8000:8000 \
  -v $(pwd)/model_repository:/models \
  nvcr.io/nvidia/tritonserver:24.01-py3 \
  tritonserver --model-repository=/models

Client Inference

uv run src/fashion_parser/scripts/triton_client.py path/to/image.jpg

---

📁 Project Map

fashion-parser/
├── src/fashion_parser/
│   ├── models/        # PyTorch model definitions & wrappers
│   ├── data/          # Torch Dataset & Loader engines
│   ├── config/        # Environment-driven settings
│   └── scripts/       # Training, Exporting & Triton clients
├── model_repository/  # Triton deployment configurations
├── scripts/           # Legacy entry points (for reference)
├── old_code/          # Backup of legacy TF/Keras implementation
└── pyproject.toml     # Modern project configuration

---

🖼️ Example Results

---

📑 Roadmap

• Phase 1: Restructure & Modernize Codebase
• Phase 2: Migrate to PyTorch & uv
• Phase 3: Triton Inference Server Integration
• Phase 4: Add Inference & Visualization Jupyter Notebooks