AI and Human Art Classifier is a web application that labels an uploaded image as either AI-GENERATED (image generated by an artificial intelligence or machine learning model) or HUMAN-CREATED (or simply an image that is not AI-GENERATED). You can try out AI and Human Art Classifier here.
AI and Human Art Classifier utilizes a fine-tuned Vision Transformer (ViT) model to classify an image from two labels: AI_GENERATED and NON_AI_GENERATED. A label will be chosen by the model based on the highest probability generated upon passing to the model. That probability will also determine the confidence level of the classification. The model is fine-tuned using 10% of Kausthub Kannan's AI and Human Art Classification Dataset to reduce training time at a cost of low classification accuracy. You can view the model at 🤗 Hub here. You can check ai_model/AI and Human Art Classifier Model.ipynb to see the model configurations. You can also check ai_model/Test Model.ipynb to test the model using a Jupyter notebook.
To be able to easily use the AI and Human Art Classifier, the model is wrapped in a custom API function to be called by a Django server. The user can upload the image they want to pass to the model and view the classification results of the model through a React client. The user is also able to provide a feedback if the prediction label set by the model is accurate to the user once the classification results have been sent to the client.