Hello,
I am a PhD student in artificial intelligence and deep learning, I had trained many models in Keras using colvolutional neural nets, and if you want I can help you to improve your accuracy, and without even GANs, I have a technique that can help you in your uneven distribution of categories problem.
So, you can pay me only if you got more than 90% accuracy.
All what I need is:
1- Your dataset.
2- An access to a powerful machine to train the models, and if the Google Colab is sufficient, I can work with it.
To be honest, I don't have big experience in TensorFlow, so if it's okay, I can do it in Keras, which is a good framework, and I have big experience on it.
Just contact me for further discussion, even if you will not hire and choose my bid .
My bests.