ISeeU

With the increasing realism of AI-generated images, it becomes increasingly difficult for one to discern whether an image is AI or not. This can lead consequences from larger-scale ones like fake events posted online that can fear-monger many to smaller, perhaps less harmful ones like a prank on one's parents. With the "Code For Change" motto in mind, we created ISeeU to allow users to upload images and check whether it is AI or not reliably. ISeeU is an AI image detector that helps in classifying whether images are real or fake.

Our solution thereby directly addresses the problem statement by building a machine learning model that is able to detect and classify real versus fake images at a consistent, good accuracy.

Development tools used (e.g. VSCode, Colab, Jupyter)

VSCode, Python

Models or APIs used

CLIP ViT Patch 14, Stable Diffusion 1.4 VAE

Libraries and frameworks used (e.g. Hugging Face Transformers, PyTorch, scikit-learn, pandas)

Used PyTorch and its children libraries mainly along with Hugging Face Transformers and scikit-learn for the required model performance metrics

Datasets and assets used

CIFAKE - train subfolder for train data && WildFake COCO val2017 & WildFAKE DALLE3 Advanced datasets for test and validation datasets

Results

Robustness Evaluation Summary

Built With

• huggingface
• machine-learning
• numpy
• python
• pytorch
• scikit-learn
• streamlit
• torchvision