This is my project done under VLG IITR. In a world where most services are available online, identity verification is crucial to ensure that only real individuals can access and use these services. However, with the rise of generative AI, fake identities can be easily created using sophisticated algorithms. This has led to an increase in identity fraud, as fake identities can be used to gain access to online services and commit fraudulent activities. To address this issue, I have developed various algorithms that can accurately distinguish between real and fake profile photos produced by generative AI. This problem is one of the competition on bitgrit and it asked to generate the csv files of the solution generated by your model for the submission. That's why I have also generated some csv files which you can find by name solution.csv, solutionbyANN.csv, soutionbyXGBoost.csv, and one other. Varioud models which I have implemented for this challenge are: XGBoost Hyper-tuned using GridSearchCV, XGBoost with Random forest, Random Forest, SVM with kernel of polynomial of dgree 2, ANN model, CNN model, and Vision Transformer. XGBoost using GridSearchCV gave the best result on submitting with f1 score of 0.78 You can assess the dataset used by following the link: https://bitgrit.net/competition/18

To know the results of the project, you can refer to Conclusion section of this readme file.

Clone this repository.

git clone https://github.com/deep0505sharma/AI_Image_Detection.git

Refer to the conclusion.txt file to see the quantitative results of the algorithms applied. If you find any discripency or issue or have any suugestions to improve the accuracy of the models, you can conatct me on Linkedin