Approach for boxing recognized image with Deep Learning

User 1241 | 2/23/2015, 6:05:58 PM

Hi, I recently experiment with the deep learning lib using the built-in algorithms mnist and imagenet to do image recognition. The result is great and so I want to move one step forward. The next step of the process is to be able to locate the position of the recognized object in the image. What would be the approach for doing that? Does graphlab deep learning has such feature?

Any guidance and advice will be deeply appreciated. Thank you.

Comments

User 940 | 2/24/2015, 6:49:40 PM

Hi @kevinchan,

It's great to hear that you are having good results with our Deep Learning toolkit! Unfortunately, we do not support bounding box detection in GraphLab Create. However, we do provide an SDK that could be used to implement your own on top of our Deep Learning models. This is an interesting <a href="http://arxiv.org/pdf/1311.2524v5.pdf">paper</a> on the subject.

Let us know if you have any more questions.

Cheers! -Piotr


User 1241 | 2/24/2015, 6:53:41 PM

Hi Piotr,

Thanks for the quick reply, the link to the Paper is broken. Can you attach the link again?

Thanks.


User 91 | 2/24/2015, 6:55:50 PM

The link has been corrected. It should be working now.