Raiden Shogun Origin Classifier👾

• Hugging Face space
• Hugging Face Repo

Inspiration

This project was inspired by the work done in the repository Detecting AI-Generated Fake Images by nogibjj. The approach and methodologies explored in this project provided valuable insights and formed a foundationd basis for the development of the Raiden Shogun Origin Classifier. We aimed to dbuild upon these idea and adapt them specifically for distinguishing AI-generated and human-created images of Raiden Shogun, showcasing the practical applications of learning in the realm of image classification.

Introduction

Raiden Shogun Origin Classifiern is a project that navigates the intersection of AI and anime. By scraping AI-generated anime images from Pixiv using the PixivBatchDownloader and comparing them with human-created ones, this project aims to train a machine learning model to recognize and differentiate these two types of artwork.

Project Components

• Pixiv Image Scraper: We harness the PixivBatchDownloader to accumulate a rich dataset of anime images, annotated based on whether they are AI-generated or human-created.
• Machine Learning Model: Using Google Colab, we develop an AI model that learns from the dataset. The model's goal is to classify images accurately into their respective origins.

Data Collection and Cleaning

Data scraping is performed via the PixivBatchDownloader, which is configured to distinguish between 'ai' and 'non-ai' tagged images. Data cleaning is an integral part of the process, ensuring that the model is trained on high-quality and relevant images only.

Visualization and Analysis

After training, visualize the model's performance using confusion matrices, ROC curves, and other relevant metrics. This will help in understanding the model's strengths and areas for improvement.

Results🔮

After training the model on the collected dataset, we achieved the following results:

• AI-Generated Artwork Classification

  • Precision: 98%
  • Recall: 90%
  • F1-score: 94%

• Human-Created Artwork Classification

  • Precision: 69%
  • Recall: 92%
  • F1-score: 79%

• Overall Model Performance

  • Accuracy: 91%
  • Weighted Avg Precision: 92%
  • Weighted Avg Recall: 91%
  • Weighted Avg F1-score: 91%

Todo

Moving forward, the project aims to:

• Increase the dataset size for better model generalization.