The future of TensorFlow-Slim vs Keras 2.0

[Bogdan Gaza]

Hello TensorFlow Community,

With the imminent release of TensorFlow 1.1.0 and the inclusion of Keras 2 API into contrib, I wonder what is the future of TensorFlow-Slim?

Personally I found really useful to see how more complex models are implemented and how we can use them with something just slightly above vanilla TF. I'm talking about the models found at: https://github.com/tensorflow/models/tree/master/slim/nets (for example: inception_resnet_v2 which is not present in the contrib package in any of the TF releases).

Also for what I can tell, the repo mentioned above also contains the official links to the inception pre-trained models that Google currently offers.

Is there any place for Slim on the tensorflow roadmap? Should we invest more time and code into using Slim or should everyone just move to Keras 2.0 once it lands on the latest release?

Thanks,

Bogdan

[Yaroslav Bulatov]

cc sguada

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+unsubscribe@tensorflow.org.
To post to this group, send email to dis...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/CAF2LzP2hyO2Zt9%3DrN7xQMPUMs9w7QguqSZD7UhLYCsTonCkbfA%40mail.gmail.com.

[Bogdan Gaza]

Bump!

[Sergio Guadarrama]

Most of TensorFlow-Slim has been merged into tf.contrib.layers, tf.contrib.losses, and tf.contrib.metrics. 

Depending on the problem a more Object Oriented approach or a Functional approach can be better suited. So now one can use either, since both share the same implementation, tf-slim follows a more functional approach while Keras follows a more OO approach,

We plan to keep releasing tf-slim models under https://github.com/tensorflow/models/tree/master/slim/.

On Thu, Apr 27, 2017 at 11:20 AM Bogdan Gaza <bc....@gmail.com> wrote:

Bump!

On Tue, Apr 25, 2017 at 12:13 PM, Yaroslav Bulatov <yaros...@gmail.com> wrote:

cc sguada

On Mon, Apr 24, 2017 at 10:55 PM, Bogdan Gaza <bc....@gmail.com> wrote:

Hello TensorFlow Community,

With the imminent release of TensorFlow 1.1.0 and the inclusion of Keras 2 API into contrib, I wonder what is the future of TensorFlow-Slim?

Personally I found really useful to see how more complex models are implemented and how we can use them with something just slightly above vanilla TF. I'm talking about the models found at: https://github.com/tensorflow/models/tree/master/slim/nets (for example: inception_resnet_v2 which is not present in the contrib package in any of the TF releases).

Also for what I can tell, the repo mentioned above also contains the official links to the inception pre-trained models that Google currently offers.

Is there any place for Slim on the tensorflow roadmap? Should we invest more time and code into using Slim or should everyone just move to Keras 2.0 once it lands on the latest release?

Thanks,

Bogdan

--

You received this message because you are subscribed to the Google Groups "Discuss" group.

To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.

[johnnyc...@gmail.com]

It's Dec 2017 (Tensorflow 1.4.1) and I've noticed we now have the official tf.layers, tf.losses and tf.metrics - are these the decedents of   tf.contrib.layers, tf.contrib.losses, and tf.contrib.metrics that you mentioned?

I notice that when I run through the slim walkthrough notebook (with Python 3.6 and Tensorflow 1.4.1) there were some warnings (about certain slim functions been removed and suggested to move on to tf.xxx instedad, such as tf.losses.xxx), hence the questions / discoveries.

Thanks!

Johnny