How can I test SegNet architecture in new caffe version with Cuda 8?

[moinulkuet]

Hi, I am new in DL. I tried to build SegNet caffe using CUDA 8 with GTX 1080 GPU card. But it seems not working, because of CUDA version mismatch. Even I can't use old CUDA version as GTX 1080 is only compatible with CUDA 8. I wonder, how can I test SegNet architecture in my existing caffe? Do you think, if I transfer all of your added files in another caffe, then it will work? Can you help please to figure out which files be needed to transfer? Thanks a lot!

[Timo-hab]

I could compile caffe-segnet using CUDA 8. You need to build it without CUDNN.

[Maxfashko]

Hello. Why not use cudnn? Why not add support cudnn in caffe-segnet?

[Timo-hab]

@Maxfashko Do you know how to add cudnn support?

[nathanin]

This is concerning for me, as well. I've built SegNet with CUDNN v3 instead of 5, and everything seems to work on the Quadro K4000 I have installed.

However, I'm now worried about upgrading my GPU to a GTX 1080. I'm not a CUDA expert; will I still be able to do this trick and compile SegNet with CUDNN v3?

[ronalddas]

@nathanin Hi, I have been trying to compile SegNet with CUDNN v3, but my caffe compilation fails. identifier "cudnnActivationDescriptor_t" is undefined I am not able to compile caffe-SegNet and BVLC-Caffe with any version of CUDNN( have tried v2,v3,v4,v5)

[alexgkendall]

FYI I've added a note to the README.md:

This version supports cudnn v2 acceleration. @TimoSaemann has a branch supporting a more recent version of Caffe (Dec 2016) with cudnn v5.1: https://github.com/TimoSaemann/caffe-segnet-cudnn5