DDColor Triton Deployment

Quick setup guide for deploying DDColor image colorization model with NVIDIA Triton Inference Server.

Prerequisites

• NVIDIA GPU (device 1 used in examples)
• Docker with GPU support
• Git and wget

Setup Steps

1. Clone Repository and Download Model

# Clone DDColor repository
git clone https://github.com/piddnad/DDColor.git

# Download pre-trained model weights
wget https://huggingface.co/piddnad/DDColor-models/resolve/main/ddcolor_paper_tiny.pth \
  -P /mnt/raid0/bharat/customer_engg/ddcolor/DDColor/pretrain

2. Model Development (PyTorch)

Run the lab tutorials to create your Triton model repository:

docker run -it --rm --gpus '"device=1"' \
  -v $PWD:/workspace \
  --net=host --ipc=host \
  nvcr.io/nvidia/pytorch:24.12-py3

3. Run the Tutorial Notebooks:

Follow the sequence

4. Deploy with Triton Server

Once the model repository is created:

# Start Triton container
docker run -it --rm --gpus '"device=1"' \
  -v $PWD:/workspace \
  --net=host --ipc=host \
  nvcr.io/nvidia/tritonserver:24.12-py3

# Install dependencies
pip install opencv-python==4.10.0.82
apt update
apt-get install ffmpeg libsm6 libxext6 -y

# Start Triton Server
tritonserver --model-repository=/workspace/model_repository_trt \
  --log-verbose=1 \
  --allow-metrics=true

Project Structure

.
├── DDColor/                    # Cloned repository
│   └── pretrain/              # Model weights
│       └── ddcolor_paper_tiny.pth
└── model_repository_trt/      # Triton model repository (created during setup)

Notes

• GPU device 1 is specified in all commands
• Host networking and IPC are enabled for optimal performance
• Metrics are enabled for monitoring
• Verbose logging helps with debugging

References

• DDColor GitHub
• DDColor Model Weights
• NVIDIA Triton Documentation