DeepClas4Bio API

deepclas4bio

#1

DeepClas4Bio is a project that aims to facilitate the interoperability of bioimaging tools with deep learning frameworks. In particular, we have developed an extensible API that provides a common access point for classification models of several deep learning frameworks. This project groups the main deep learning frameworks, namely, Keras, Caffe, DeepLearning4J, MxNet and PyTorch.

Using DeepClas4Bio, the users could work with the main deep learning frameworks easily and transparently for them. The users can employ the pretrained models included in the API for classify an image or they can train their own models and load them in the API to use them in a simple way.

For more information you can visit the project webpage or contact with me in the email adrian.ines@unirioja.es.


#2

Hi, It is cool, but it is a command line interface, can you give a python interface, and support pip install? you use subprocess everywhere, and if user use a virtual, subprocess(‘python’ …) will not work. Can you give a python api, we can use like below:

import deepclas4bio as dc4b

frames = dc4b.listFrameworks(…)
models = dc4b.listModels(…)

rst = dc4b.predict(img, …)

I also see you write a ImagePy plugin, may be it is not the best usage, I will help to improve, but I think we need a python api first.


#3

Hi, it is a good idea. Our first idea was to make an API that would allow us to connect with any programming language, because of this we decided to use the command line interface. Your suggestion is interesting and we will give a python API soon.

Thank you.


#4

my advice is support pip install, and give a python api. (easy to install and easy to use, elegant and powerful, that is why python is so popular). we should serve python user first.

then you can use python -m xxx [parameter] in other language, not python the absolute file. (And I also remind there is some package can make a python package as a command line tool easily, you can try to search)

if so, I think the imagepy plugin could be written in very short code (the lion demo should be in about 10-20 line, I will help to improve).


#5

Our idea is to give a python API that supports pip install. Also, as you have mentioned, we want to change the calls to the API from other programs, using this pip support (making the code more elegant and avoiding calls to python files). We hope to have some progress next week.

Thanks for your suggestions and advices.