Cntk: Pixelwise classification using CNTK

[rold2007]

Hello,

I would like to use CNTK for pixelwise classification of images (like SegNet). I haven't found a ready-to-use network architecture in the examples folder to do such a task. I have used AlexNet's model for binary classification of images successfully but I am not sure where to start to make this kind of network evolve to generate a pixelwise output.
Could anyone point me in the right direction ?
Is it a good idea to start from AlexNet for this ?
In case it is relevant, I am currently only doing binary classification, but I would be interested in a solution with also (potentially) supports multi-class.

Also, from what I could understand, to generate the pixelwise output I'll need to feed the ground truth (which I don't have right now) of the same size as my training images in input. Creating the ground truth of 10k+ images is a huge task, is there a way to produce pixelwise output without a pixelwise ground truth as input ?
If the ground truth images are compulsory, for a start I could do some tests using the CamVid or KITTI datasets but I would be happy if I wouldn't have to produce all those ground truth images...

Thanks !

[cha-zhang]

We are preparing examples for pixelwise classification (segmentation), please stay tuned.

[rold2007]

@cha-zhang Thanks a lot ! Is there any timeframe, 1 week, 1 month ?

[cha-zhang]

No firm timeframe yet, but definitely longer than 1 week. We need to prepare a few other examples before moving to the segmentation task.

[pksorensen]

@cha-zhang Any update on this? :)

[cha-zhang]

Not yet. It’s holiday season and everything moves slowly…

From: Poul Kjeldager Sørensen [mailto:[email protected]]
Sent: Monday, December 19, 2016 1:54 AM
To: Microsoft/CNTK [e-mail]
Cc: Cha Zhang [e-mail]; Mention [e-mail]
Subject: Re: [Microsoft/CNTK] Pixelwise classification using CNTK (#933)

@cha-zhanghttps://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fcha-zhang&data=02%7C01%7Cchazhang%40microsoft.com%7C86032dde141440f3397108d427f500c3%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636177380607644506&sdata=nFivCjOiDZGVy5quuonx9ACqrt6z4ztVoyZ63n1xCVA%3D&reserved=0 Any update on this? :)

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHubhttps://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FMicrosoft%2FCNTK%2Fissues%2F933%23issuecomment-267924113&data=02%7C01%7Cchazhang%40microsoft.com%7C86032dde141440f3397108d427f500c3%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636177380607644506&sdata=YnOLAfcSK90KzwDpJQ2H%2Fyy25LB%2FoqACxo4liK%2BH2wE%3D&reserved=0, or mute the threadhttps://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FARnjacCqmB-s7aIKs57TQBhYEaTMr20nks5rJlRKgaJpZM4KdyKE&data=02%7C01%7Cchazhang%40microsoft.com%7C86032dde141440f3397108d427f500c3%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636177380607644506&sdata=HAlka6DqRzZGdN7kJPubJ9qFcCx5YBLifYGbXI30h3g%3D&reserved=0.

[dimadim]

@cha-zhang Can CNTK do pixelwise classification of images like SegNet now? Is there any examples? Thank you )

[cha-zhang]

The key components should have already been there. We are still working on the examples.

[pksorensen]

An example like SegNet would be cool :)

[matiasmolinas]

Any update?

[ldenoue]

Anything now?