🖌️ DefectDetect-Application

A lightweight tool for annotation and patch generation, originally developed for defect marking on leather surfaces, but easily adaptable for general image annotation needs.

📌 This work was partially supported by the Federal Ministry of Science and Education of Bosnia and Herzegovina under the project "Automated Visual Inspection Based on Deep Neural Networks" in 2023/2024, at Faculty of Electrical Engineering, University of Sarajevo.

---

🔎 Overview

DefectDetect enables manual image annotation via freehand drawing, the creation of smaller image segments called patches, and flexible export of annotation data in structured JSON format.

Key features:

• 🧩 Patch creation & export with defect severity ratings (0–2)
• 📁 Export in both individual and collective JSON files
• ✅ No installation required — standalone executable

---

📁 Project Structure

DefectDetect-Application/
├── data/                 # Input test images
├── instructions/         # Detailed instructions on how to use the app
├── notebooks/            # Jupyter analysis/evaluation
├── qt_runtime/           # Application source code
├── README.md             # Project description
└── .gitignore

---

📥 Download

🪟 For Windows App: The application does not require installation or dependency setup.

1. Go to the Releases section.
2. Download the latest release .zip package.
3. Unzip and run the executable file directly.

✅ That’s it — no Python, no installation, just run and annotate!

🐧 For Linux App

1. Go to the Releases section.

2. Download the latest '.zip' package for Linux where you can find DefectDetect-x86_64.AppImage file (a single-file Linux executable).

3. Make the file executable and run it:

   chmod +x DefectDetect-x86_64.AppImage
   ./DefectDetect-x86_64.AppImage
   

---

📖 How to Use

For detailed instructions:

• 🪟 Windows usage guide
• 🐧 Linux usage guide

---

📦 Export Options

• Patch-based Export: Create and save patches as separate images.
• Ratings: Assign 0 (none), 1 (partial), or 2 (full) for defect presence in each patch.
• JSON Files:
  • Individual: One JSON file per patch
  • Collective: One JSON file containing all patches, masks, ratings, etc.

---

📄 JSON Structure

Each collective JSON file includes:

• annotations: patch ID, ratings, mask links
• classes: list of defect classes
• masks: masks per patch
• ratings: meaning of each rating
• org_patches: position & size of patches

For code examples on parsing and using the JSON files in Python (e.g. in Colab), see notebooks/DefectDetect_JSON_Examples.ipynb.

---

🧠 Improvements Planned

• Support for rectangle/circle annotations
• GUI enhancements & internationalization

---

📚 Citation

If you use this tool in your research or publications, please cite:

Arapović, L. (2024). DefectDetect: Application for Defect Annotation on Leather Surfaces. University of Sarajevo – Faculty of Electrical Engineering. (to be changed)

---

📬 Contact

Developed by: Lejla Arapović
Email: [email protected]
Faculty of Electrical Engineering, University of Sarajevo