New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Prebuilt models for image classification #140

Closed
shethsh opened this Issue Feb 19, 2016 · 4 comments

Comments

Projects
None yet
4 participants
@shethsh

shethsh commented Feb 19, 2016

I am transitioning from Caffe and planning to use pre-built models for finetuning in CNTK. I am not able to find a model repository. Can somebody provide pre-built models for image classification?

@Alexey-Kamenev

This comment has been minimized.

Alexey-Kamenev commented Feb 20, 2016

Hello,
We currently do not have pre-trained models available on GitHub. This is certainly something we plan to do, hopefully very soon.

@shethsh

This comment has been minimized.

shethsh commented Mar 1, 2016

Hi,
Thanks for the response.
I wanted to know if we can stop and resume training a network.
I saw "adapt", but, it seems that we need to specify a ref node name and seems to be suitable only for Finetuning. But, if I want to train in parts, say, whenever I get new data, I will train and store the model. How can this be done?
Also, it will be great if you can provide more info on "adapt", regarding where it find applications.

@frankseide

This comment has been minimized.

Member

frankseide commented Mar 1, 2016

Yes, you can. CNTK keeps checkpoints. You can always pick up a training after a completed epoch.

If this does not happen automatically, please say “makeMode = true” somewhere in the outer level of the main config (and make a backup of your model folder before you try this for the first time, just in case).

From: shethsh [mailto:notifications@github.com]
Sent: Tuesday, March 1, 2016 1:49
To: Microsoft/CNTK CNTK@noreply.github.com
Subject: Re: [CNTK] Prebuilt models for image classification (#140)

Hi,
Thanks for the response.
I wanted to know if we can stop and resume training a network.
I saw "adapt", but, it seems that we need to specify a ref node name and seems to be suitable only for Finetuning. But, if I want to train in parts, say, whenever I get new data, I will train and store the model. How can this be done?
Also, it will be great if you can provide more info on "adapt".


Reply to this email directly or view it on GitHubhttps://github.com//issues/140#issuecomment-190638323.

@shethsh

This comment has been minimized.

shethsh commented Mar 5, 2016

Thanks. I am able to resume training from Epochs. Closing this issue

@shethsh shethsh closed this Mar 5, 2016

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment