deeplab v2 pretrained model doesn't work

24 views
Skip to first unread message

weiyi.z...@gmail.com

unread,
Jul 25, 2017, 11:25:52 PM7/25/17
to Caffe Users
  I just want to test the pretrained model given by deeplab v2. So I write a deploy.prototxt by myself. But I find that the model doesn't work when I remove the DenseCRF layer. And with the DenseCRF layer, the outline is clear but the corresponding label change always, that is, the model can't really predict.

  Is there anything wrong with my deploy.prototxt ? I have used the "train_iter_20000.caffemodel"  of DeepLabv2 (VGG-16) downloaded from http://liangchiehchen.com/projects/DeepLabv2_vgg.html. I think it has been already trained on PASCAL VOC 2012 data set. So I set the NUM_LABELS to 21 in my deploy.prototxt.

  Can anyone help me? THANKS A LOT!
deployv2.prototxt
Reply all
Reply to author
Forward
0 new messages