In the future, they should expand text processing, for a recommendation system, or to support some other models as well — that would be great. The concept of Caffe is a little bit complex because it was developed and based in C++. They need to make it easier for a new developer, data scientist, or a new machine or deep learning engineer to understand it. You can't work with metrics and vectors as Python does.

Python is a vector-oriented language, but Caffe is not. When you deal with memory in C++, you have to allocate the data you will use in memory. You have to manage everything in C++. Conversely, in Python, you don't need to do that since everything is abstract and done by Python itself. It depends on every use case or your requirement goals. Some clients will require you to use Caffe because maybe their projects are old and they want to continue with Caffe. Others are comfortable with their current situation or they are afraid of migrating to another library. From my point of view, they need to make it easier for a new developer to use it. They should incorporate Python API to make it richer, overall.