Keras

What you need definitely to start your deep learning experiments

Commentaires : I would defintely recommend it as the quickest step to start testing your model.

Avantages :

Keras is the only platform that runs on top of most popular backends like TensorFlow, pyTorch and Microsoft Cogntitive Toolkit. This gives great flexibility to researchers to try their network architecture with minimal changes across multiple libraries mentioned. The sequencing modularity is what makes you build sophisticated network with improved code readability .

Inconvénients :

If you encounter an error, it is hard to be debugged.

Keras for deep learning

Commentaires : I did many deep learning projects using keras it is really helpful

Avantages :

easy to use, large communities and support

Inconvénients :

keras has many predefined methods and functions but it is difficult to integrate a custom class.

Keras for school project

Avantages :

I did use this library couple of times during the semester to solve my deep learning course home works and project. compared to tensor flow it was easier for me to use

Inconvénients :

It was not still easy to use and well documented with examples

Great Deeplearning framework

Commentaires : i use keras for image classification making use of it's pretrained architectures especially the resnet architectures.

Avantages :

What i love most about keras is it's wrapper functions, i use it to perform Gridsearch using scikitlearn and this is amazing as i cannot do this on other frameworks. keras also has a good documentation page with lots of pretrained CNN architectures for image classifications solutions.

Inconvénients :

Nothing to dislike about this framework yet.

Start Learning From Keras Framework

Commentaires : I recommend it for performing image classification as it provides some inbuilt fucntionality for image preprocessing. It even comes with many usefull pre-trained models like resnet.

Avantages :

First thing i like about Keras is that it runs on the top of tensorflow background. Deep learning and neural network construction and visulaization is simple using Keras, also it comes with enough documentations. It provides lots of inbuilt functions for image processing which makes it lots easier for image classificaiton.

Inconvénients :

For building more customized deep learning model, you need to use TensorFlow. Also the model inferencing time is little slow compared to model directly build in TensorFlow.