Weebo NSFW: Image Classification Project

This is an Image classification project for detecting NSFW image/content based on YOLOv8. It also uses TensorFlow 2 to classify drawing images (anime, waifu games) versus real pictures (people, animals, etc.). This project serves as a personal exploration into object detection and image classification techniques.

Overview

• NSFW Detection: Utilizes a custom-trained YOLOv8 model to detect and classify NSFW content with bounding boxes.
• Image Type Classification: Uses a TensorFlow 2 model to distinguish between anime-style/drawn images and real-life photos.
• Automated Sorting: Automatically processes images placed in a designated test folder and outputs the classified results into categorized directories.

How It Works

1. The project uses a TensorFlow 2 model to first determine if an image is anime/drawn or a real-life picture.
2. It then runs a YOLOv8 model to detect and label NSFW content.
3. Processed images are saved with bounding boxes, alongside .txt files containing the YOLO detection results.

You can view the YOLO model training report on Weights & Biases.

Getting Started

To try out the project, you can find the source code, pre-trained models, and installation instructions on the GitHub repository.

Make sure you have Python 3.8+ installed, then install the required packages:

pip install tensorflow ultralytics

Note: The model was not trained on Furry content, so predictions for such content may not be accurate.