Much like FastAI for pytorch, Tensorflow provides native abstraction layer with keras, which simplify the task of creating convolutional neural network model.

But why tensorflow? you may ask. Haven’t we been learning about FastAI? Personal preferance… With Tensorflow, you can convert the model into Tensorflow lite, hex dump it and run it in your favourite micro-controller like ESP32.

With intuitive official documentation, AI detection model was created with minimalistic approach, which yielded close to 95% accuracy rate for detecting AI images. Fantastic. Could I do better?

Yes… -data augmentation -epochs -over-trained? -What about bigger images?

To be continued…