🚫 BlurScene — Privacy-preserving Anonymization for Traffic Data

BlurScene is a deep learning model designed to anonymize faces and license plates in traffic-related images and video data.

---

📋 Table of Contents

1. 🔧 Setup
   • 📦 Environment
2. 🧠 Inference
   • ⚙️ Configuration
   • 📥 Model Weights Download
   • 🧪 Inference Script
   • 🌐 Server
   • 🐳 Docker

---

🔧 Setup

📦 Environment Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

✅ Requires Python ≥ 3.10 (tested with 3.10.12)

---

🧠 Inference

⚙️ Configuration

All inference settings can be found in config/inference.yaml.
You must provide:

• ✅ Path to the model weights .pt
• ✅ Path to the model configuration .yaml (Hydra config)

📁 It's recommended to store both in a weights/ folder (which is whitelisted in .dockerignore).

🧰 Optional Processing Steps:

• mirror_image: horizontal mirroring
• enlarged_regions_n: crops the image into n × n subregions

🔎 Post-Processing Parameters:

• pre_merge_score_threshold, post_merge_score_threshold: filter weak predictions
• merging_method:
  • wbf: Weighted Box Fusion
  • nms: Non-Maximum Suppression
  • nmm: Non-Maximum Merging
• merge_iou_threshold: IoU threshold for merging
• area_method: how box area is computed:
  • int: (x1 - x0 + 1) * (y1 - y0 + 1)
  • float: (x1 - x0) * (y1 - y0)

You can also configure logging level and format via the YAML.

---

📥 Model Weights Download

You can download the pretrained model weights and matching configuration from the following link:

👉 Download BlurScene Weights

📁 Place the downloaded files inside the weights/ directory.
⚠️ Make sure the .pt and .yaml files match exactly — otherwise inference may fail.

---

🧪 Inference Script

You can test inference directly via:

python inference.py path/to/image.jpg

This loads the model, compiles it if necessary, and outputs detections for the image.

---

🌐 Server

Start a local Flask server using:

python run_server.py

Or use Gunicorn (recommended):

gunicorn --config=config/gunicorn.py run_server:app

🔧 Gunicorn reads environment variables:

• WORKERS
• WORKER_TIMEOUT
• PORT
• LOG_LEVEL

Default address is http://0.0.0.0:5000.

🧪 Test the Endpoint

curl -H "Content-Type: image/jpeg" --data-binary @test.jpg http://localhost:5000/anonymize --output returned_image.jpg

---

🐳 Docker

🏗 Build

docker build . -t blurscene

Note: Settings in inference.yaml (e.g. device) are fixed at build time.

🚀 Run

docker run -d --gpus all -p 5000:5000 --name blurscene_container blurscene

🔍 Monitor logs:

docker logs -f blurscene_container

Wait for model compilation to complete before sending requests.

---

📑 Citation

@misc{vosshans2024aeifdatacollectiondataset,
    author    = {Marcel Vosshans and Alexander Baumann and Matthias Drueppel and Omar Ait-Aider and Ralf Woerner and Youcef Mezouar and Thao Dang and Markus Enzweiler},
    title     = {The AEIF Data Collection: A Dataset for Infrastructure-Supported Perception Research with Focus on Public Transportation},
    url       = {https://arxiv.org/abs/2407.08261},
    year      = {2024},
}

📜 License

This project is released under the MIT License.