On initial research, I found that that using ResNet would give us higher accuracy and many research papers explain why ResNet is very efficient to classify AI generated fake images, I would be working on implementing the ResNet, first by using the already trained weights and performing transfer learning on the same with our dataset that we create. Once we have a good accuracy , we would further consider other architectures like MobileVNet or further fine tune the same model.
On initial research, I found that that using ResNet would give us higher accuracy and many research papers explain why ResNet is very efficient to classify AI generated fake images, I would be working on implementing the ResNet, first by using the already trained weights and performing transfer learning on the same with our dataset that we create. Once we have a good accuracy , we would further consider other architectures like MobileVNet or further fine tune the same model.