mark

Have you changed log layer? I see loglayer diffs in commits.

I just added a loss layer called triplet_loss_layer and defined a parameter of this layer called 'margin' in upgrade_proto, other changes are all obeyed on the rules on Caffe new layer contribute instruction.
By the way, my whole code is working well now, an example on MNIST dataset with the format of LMDB is also included in examples/triplet folder. When I finished the GPU version of this triplet loss layer, all build will be passed. @bhack @mtamburrano

I still see log layers in the diffs.

Yes, I do see the diffs in log_layer, it's caused by the version updating from a newly released version, some files which I don't care in details changes.
The macro definition is always changing as new layer added and I must change some macro definition of triplet loss layer as the index being used by new accepted layer, how to deal with it as the triplet layer not merged in Caffe? @bhack

You need to rebase regularly. See in particular the section "When you are ready to develop a feature or fixing a bug, follow this protocol"

@pwohlhart My code for an easier loss function which produce the similar result as described in paper[1] is done in mnist database, but as I have also rendered some images using OpenCV in LINE-MOD database, the tranining samples are not easily arranged, will your codes containing the data generation process?

@FlorisGaisser Your concern about the label could be solved in this paper~

http://lrs.icg.tugraz.at/pubs/wohlhart_cvpr15.pdf

It is reopened as #3103

Now at #3123

Thanks for the great work. I met a problem with the installation.
I succeed in $make all and $make test, but failed in $make runtest (The 4 tests of TripletLossLayerTest all failed). Could anyone help?