multi-label CNN implementation using many of loss function

[Samer Iskander]

Hello:

In the paper "Pearson Attribute Recognition with a Jointly-trained Holistic CNN Model", the authors proposed to fine-tune the CaffeNet (similar to AlexNet, but exchange two layers). They converted the problem from multi-class prediction to multi-label prediction. That means the image may have one or more attribute (one or more label) at the same time. Each label (attribute) has a separate loss function. 

How can I implement that using torch7? How to construct that CNN?

Thank you.    

[Beanfrog Green]

1.load caffenet: loadcaffe
2.multi branch: nn.ConcatTable
3.multi loss: nn.ParallelCriterion

see more: https://groups.google.com/forum/#!topic/torch7/OLjblK6iVl0

在 2016年4月13日星期三 UTC+8上午10:43:04，Samer Iskander写道：