This repository contains the source code for the semantic image segmentation method described in the ICCV 2015 paper: Conditional Random Fields as Recurrent Neural Networks. http://crfasrnn.torr.vision/
Other
1.34k
stars
462
forks
source link
compiling with Cuda 8.0 support, but not using GPU during train #78
The training seems to be using CPU only and takes forever. There are a few MB used on the GPU, but no computational load done.
Also the MultiStageMeanfieldLayerTest/x.TestGradient (make runtest for bittnt/caffe) is not using computing capabilities of the GPUs, other caffe-tests do.
Am I completely overlooking something? How can I make crfasrnn use a cuda-8.0 GPU?
many thanks in advance!
Hi,
Technically I can compile with Cuda8.0-Support (USE_CUDNN := 1), but the GPU seems not to be used.
details: I got a Pascal-Chipset based cuda-8.0 GPU (GTX 1080) and can compile https://github.com/bittnt/caffe (e.g. ab4118d) Then I use this caffe version for training the examples given in https://github.com/MasazI/crfasrnn-training or
https://github.com/martinkersner/train-CRF-RNN (I made the adaption necessary to the prototxts)
The training seems to be using CPU only and takes forever. There are a few MB used on the GPU, but no computational load done.
Also the MultiStageMeanfieldLayerTest/x.TestGradient (make runtest for bittnt/caffe) is not using computing capabilities of the GPUs, other caffe-tests do.
Am I completely overlooking something? How can I make crfasrnn use a cuda-8.0 GPU? many thanks in advance!