Logo detection system with Tensorflow

We need a deep learning application that can detect brand logos on documents/images. We are looking at about 200 brands that need to be classified by the brand name: eg. pepsi. We have been doing some research and think that our best option is to go with Tensorflow, but you can also make suggestions of what is the best platform to accomplish our goals. We actually tried a couple of github demos out there. We are a team that is currently working with backend services with java and angular on AWS. We don't have much python experience but we can handle basic command line stuff (e.g. training models) The expected application should be a basic service where we can post an image or document in png format for instance. You can suggest the best format to work with in order to obtain better accuracy in less time. The result of the query will be the brand name and accuracy if found. Also, we need a simple way to re-train our models in order to add more brand logos ourselves and make it "smarter". For example, LabelImg works and we can use it, but to build the training and test files can be a big hassle. We need a simpler more automated way to retrain our system. So basically the deliverables are: - A service that receives an image and returns the detection result, we can provide a set of images to create a model - A manual step by step on how to add more images and train our model, and all the necessary steps to make it go live on AWS