• https://colab.research.google.com/github/komorra/ACA/blob/main/komorra_ACA_Augmented_Classifier_Attack.ipynb

ACA

Augmented Classifier Attack

What is ACA?

• Generative algorithm, that can generate images, with support of any image classifier
• It doesn't need image generators
• It optimizes input image space "directly", optimizing it using augmentation, to match given classifier output
• It can help researchers to optimize their image classifiers

It is highly recommended to run this Colab notebook with GPU backend

Examples

Example using OpenAI CLIP model

Each image was generated for optimizing input image for given prompt embedding vector

Prompt: Beautiful flowers

Prompt: Dance party

Prompt: Portrait of Mario

Prompt: Singing person