NSFW Detector & Annotator

A Streamlit-based web application for detecting, classifying, and blurring explicit content in images using YOLO models.

---

Table of Contents

• Overview
• Features
• Project Structure
• Installation
• Usage
• Models
• Results
• Future Plans
• License
• Contributing
• Citations
• Contact

---

Overview

The NSFW Detector & Annotator is a tool designed to classify images into categories such as drawing, hentai, normal, porn, and sexy. It also detects explicit regions in images and allows users to blur them for moderation purposes. The app is built using Streamlit and leverages YOLO (You Only Look Once) models for classification and segmentation. Support for videos has been added wherein the user can upload a videa and the model will process it and return the annotated video once done.

---

Features

• Image Classification: Classifies images into 5 categories: drawing, hentai, normal, porn, and sexy.
• Explicit Region Detection: Identifies explicit regions within images by utilizing a segmentation model trained to detect areas such as breast, anus, female_genital, and male_genital.
• Blurring: Allows users to blur detected explicit regions with adjustable blur intensity.
• Video Moderation: Identifies and blurs explicit regions in each frame.
• Interactive UI: Built with Streamlit for a user-friendly and interactive experience.

---

Project Structure

The project is organized as follows:

nsfw-detector-annotator/
├── models/
│   ├── classification_model.pt
│   └── segmentation_model.pt
├── scripts/
│   ├── train_classification.py
│   └── train_segmentation.py
├── assets/
│   ├── classification/
│   ├── confusion_matrix.png
│   └── confusion_matrix_normalized.png
│   └── segmentation/
│   ├── confusion_matrix.png
│   └── confusion_matrix_normalized.png
├── app/
│   ├── app.py
│   ├── admin.py
│   └── utils.py
├── requirements.txt
└── README.md

---

Installation

1. Clone the Repository:

   git clone https://github.com/Forenche/nsfw_detector_annotator.git
   cd nsfw_detector_annotator
   

2. Set Up a Virtual Environment (Optional but Recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install Dependencies:

   pip install -r requirements.txt
   

---

Usage

1. Run the Streamlit App:

   streamlit run app/app.py
   

2. Upload Image(s) or Videos:

   • Use the file uploader to upload a batch of images or a single image.
   • Switch to video mode and use the file uploader to upload a video to blur explicit regions.

3. View Results:

   • The app will classify the image and display the result.
   • If the image is classified as porn or hentai, explicit regions will be detected and can be blurred.
   • Use the "Blur NSFW Regions" checkbox to toggle blurring on the detected regions.
   • The video will have the explicit regions blurred out once processed.

4. Adjust Blur Intensity:

   • Use the slider to adjust the blur intensity.

---

Models

The app uses two YOLO models:

1. Classification Model (classification_model.pt):

   • Classifies images into one of the 5 classes.

2. Segmentation Model (segmentation_model.pt):

   • Detects, segments and blurs explicit regions in images.

---

Results

For the classification model:

Confusion Matrix	Normalized Confusion Matrix

For the segmentation model:

Confusion Matrix	Normalized Confusion Matrix

---

Future Plans

• Model enhancements:

  • Higher accuracy for segmentation and better generalization for classification model.

• 🔨 Real Time Video classification and Segmentation [WIP]:

  • Support for videos and classify it frame by frame in real time and segment explicit regions without dropping frames.

• Move away from streamlit to custom website:

  • Streamlit is really nice but I want to have a custom website where I have full control, freedom and access to every element.

---

License

This project is licensed under the MIT License. See LICENSE for details.

---

Contributing

Contributions are more than welcome! How to contribute?

1. Fork the Repository:

   • Click the "Fork" button at the top right of this page.

2. Clone Your Fork:

   git clone https://github.com/<your-username>/nsfw_detector_annotator.git
   

3. Create a branch:

   git checkout -b feature/your-feature-name
   

4. Make changes and commit:

   git add .
   git commit -m "Add your feature"
   

5. Push to Your Fork:

   git push origin feature/your-feature-name
   

6. Open a Pull Request:

   • Come back here and open a pull request.

---

Citations

LSPD dataset:

@article{phanlspd,
    title={LSPD: A Large-Scale Pornographic Dataset for Detection and Classification},
    author={Phan, Dinh Duy and Nguyen, Thanh Thien and Nguyen, Quang Huy and Tran, Hoang Loc and Nguyen, Khac Ngoc Khoi and Vu, Duc Lung},
    journal = {International Journal of Intelligent Engineering and Systems},
    publisher = {Intelligent Networks and Systems Society},
    volume = {15},
    issue = {1},
    pages = {198--231}
}

Model used (YOLO11):

@software{yolo11_ultralytics,
    author = {Glenn Jocher and Jing Qiu},
    title = {Ultralytics YOLO11},
    version = {11.0.0},
    year = {2024},
    url = {https://github.com/ultralytics/ultralytics},
    orcid = {0000-0001-5950-6979, 0000-0002-7603-6750, 0000-0003-3783-7069},
    license = {AGPL-3.0}
}

---

Contact

For questions or feedback, feel free to reach out to me:

GitHub: Forenche

Email: [email protected]

---