TF-Slim: Please improve installation method and documentation.

[Hvass-Labs]

I don't know if this is the right place to write about TF-Slim.

I wanted to use TF-Slim and do a Youtube tutorial on it. But it looks like it's still in the alpha-stage so I've decided to wait several months and come back and look at it again later.

I actually didn't get very far with TF-Slim before it started becoming frustrating. Here's a few comments that I hope might be useful when improving and streamlining TF-Slim:

• Installation must be simpler. It should not require installing the latest nightly build of TensorFlow. And it should not require separate installation of the image models. This is quite messy. PrettyTensor just needs: pip install prettytensor
• There seems to be a lot of overlap and redundancy in the README-files. This is very confusing and I don't know which one is correct. See for example:

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/slim

https://github.com/tensorflow/models/tree/master/slim

https://github.com/tensorflow/models/tree/master/inception/inception/slim

At this point I stopped looking at TF-Slim because it indicates that it's in the alpha-stage, so I'd rather move on to something else. Hopefully you'll improve TF-Slim in the coming months if it's intended to be a permanent part of TensorFlow that people should know how to use.

Thanks!

[jart]

Thank you for your patience @Hvass-Labs. As noted in tensorflow/tensorflow#4964, the software in the contrib directory comes with an implicit caveat emptor and is therefore only lightly supported. Whether or not that stuff becomes a permanent part of TensorFlow Core is up to the community.

If there's one thing of which I can assure you, it's that we will put in the extra time and effort to give incredible amazing code reviews to any member of the community who is generous enough to send us a pull request helping to improve these packages.

So I'm going to mark this issue as contributions welcome, and I want to thank you for bringing this to our attention.

[mingxingtan]

I would recommend using the new garden models: https://github.com/tensorflow/models/, since we are moving from slim to garden models.

I am closing this bug now. If you have similar or other issues with the new official models, please feel free to file a separate bug. Thanks!

[AbhimanyuAryan]

@mingxingtan what are garden models?