Models: Release trained im2txt model?

[mabirck]

Please let us know which model this issue is about (specify the top-level directory)

Someone could put the full trained network here, with full I mean, caption, the lstm and cnn trained together? It would be good to who does not have computational structure.

Thank you all.

[tatatodd]

@MarcoBirck I'm sorry, I don't understand. Is there a particular model that you're referring to?

[hebenon]

Without putting words in @MarcoBirck 's mouth, I think he's referring to the "Show and Tell" model im2txt.

The training time of the network is significant enough that it's a barrier to casual experimentation or proof of concept dev work. It'd definitely be helpful to have a pretrained checkpoint for the model available in the same way that the TF-slim model checkpoints are made available.

Edit: Damn you, autocorrect.

[mabirck]

Sorry I made the question without explaining it, @hebenon is right, the model of im2txt is so expensive computationally and I was wondering if you could made it available.

Thank's

[tatatodd]

@cshallue might have some thoughts on whether it's feasible to release a trained im2txt model.

[cshallue]

Hi all, thanks for the interest. Unfortunately, we're not releasing a trained version of this model at this time.

[tatatodd]

Thanks for the info @cshallue.

Closing this out, but @hebenon and @MarcoBirck feel free to update the issue (or file a new issue) if you have other problems.

[mabirck]

thank you anyway.

[ZER-0-NE]

Unfortunately, we're not releasing a trained version of this model at this time.

Still in 2019?