Was fcn-alexnet-pascal.caffemodel trained on PASCAL VOC?

[Greg Heinrich]

Hello,
I am sorry I couldn't find the answer in the forum. I am wondering if http://dl.caffe.berkeleyvision.org/fcn-alexnet-pascal.caffemodel from fcn.berkeleyvision.org had been trained on PASCAL VOC 2011 or whether this is "just" the standard Alexnet from e.g. https://github.com/BVLC/caffe/tree/rc3/models/bvlc_alexnet with the extra net surgery to transplant weights from layers fc6 and fc7 into their conv layer counterparts in the corresponding FCN model?

I hope the questions makes sense. Thanks in advance.

[Evan Shelhamer]

It was fine-tuned from ILSVRC pre-training for semantic segmentation on PASCAL VOC (and its SBD extension) as described in the linked papers at fcn.berkeleyvision.org.

Evan Shelhamer

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/c0a72d2d-f194-4ceb-99a2-3305636e8988%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Greg Heinrich]

Thank you.

[linzertling]

Hi,

    As we know, fcn is fine-tuned from vgg. But I found that in fc6 and fc7 layers, fcn has the same name with vgg and they have different types(conv and fc). How does it works?

在 2016年7月30日星期六 UTC+8上午9:04:49，Evan Shelhamer写道：

[Greg Heinrich]

Hello linzertling, see this post from Evan: https://groups.google.com/d/msg/caffe-users/rYihuHF4LFs/-R4F1ddQAwAJ