If you want to quickly build and test a neural network with minimal lines of code, Keras is the best framework. With Keras, you can build simple or very complex neural networks within a few minutes. Keras is more user friendly. Review collected by and hosted on G2.com.

If someone wants more control over your network and want to watch closely what happens with the network over the time, Tensor Flow is the right choice. Keras has been integrated in TF, it is wiser to build your network using tf.keras and insert anything you want in the network using pure TensorFlow. Review collected by and hosted on G2.com.

To understand deep learning training and layer management, keras helped me a lot. Very easy to manage, change, manipulate the layers. Parameters are also easy to tune. Review collected by and hosted on G2.com.

There are some fairly advanced frameworks out there compared to keras. It definitely helps a beginner. But I wish it could be used in a more advanced level as well. Review collected by and hosted on G2.com.

It is very easy to prototype new research ideas using Keras. The support for multi GPU and distributed training is very good. Review collected by and hosted on G2.com.

Not native to python, so makes it slightly hard to debug sometimes. Different backends are not robust often. Review collected by and hosted on G2.com.

Easy to whip up a small network without worrying about dimensionality matching Review collected by and hosted on G2.com.

Not enough control over graphs/loss functions/implementations Review collected by and hosted on G2.com.

Sprint model creation, easy to create and test simple models, automatic early stoppi g and easy to use data generators. Review collected by and hosted on G2.com.

Model transfer use from multi gpu to single gpu Review collected by and hosted on G2.com.

The best thing that I like about Keras is that it is easy and convenient to use. It is a Unified platform Review collected by and hosted on G2.com.

I am still exploring it more, will know after a few Review collected by and hosted on G2.com.

Keras drastically simplifies building neural networks in Tensorflow without losing too much customizability/functionality. I prefer Keras over TF always unless I know for sure that the best way to use some aspect is within TF itself. Review collected by and hosted on G2.com.

For beginners, bypassing concepts like matrix multiplication/dimensions can make some errors hard to rectify. So while it's a great way to get your feet wet building NNs, some coding with Tensorflow can be useful to keep your shapes in line. Review collected by and hosted on G2.com.

Model building using Keras is super easy for beginners. The fact that it uses TensorFlow as it’s backend for model building and execution is interesting because it makes the model efficient. Review collected by and hosted on G2.com.

There is nothing that I currently dislike about Keras. Review collected by and hosted on G2.com.

Easy to learn with many options and lots of examples. You can quickly find test many models or even different approaches very quickly Review collected by and hosted on G2.com.

Hard to find documentation on some of the more obscure options that are dependent on other frameworks. Review collected by and hosted on G2.com.

I really enjoy working in Keras. Its fast and efficient and is easily doable. Review collected by and hosted on G2.com.

Keras layers can only return a single tensor. For building a moderately complex network, this can result is lot of extra code which can be time consuming Review collected by and hosted on G2.com.