David Held
David Held
Have you modified vot.h at all? Lines 48 and 49 should prevent line 50 from being executed unless you have trax.h: #ifdef __has_include # if __has_include("trax.h") Strangely, it seems that...
@charliememory - this is great, nice job figuring all of this out! Any chance that you can create a quick pull request with these changes?
At the moment, yes! On Wed, Dec 20, 2017 at 8:40 PM, ekcheng wrote: > Based on this sentence in the paper, do I understand correctly that the > size...
Not sure, unless the other threads are also using the GPU, then sharing the GPU between other threads might cause a slow-down. Not sure though.
No idea - it might be helpful to dig in further. If this is a Caffe issue then you could try posting to the Caffe users group.
1) I use the pre-trained CaffeNet architecture, which is available from Caffe: http://caffe.berkeleyvision.org/model_zoo.html 2) I think it probably will - if you do this then I recommend using a much...
It sounds like you are overfitting. Just to be sure - I don't train the conv layers at all, those are pre-trained using CaffeNet.
How do you create the pre-trained network?
That's odd, not sure.
The oscillation is normal and simply occurs because the training evaluation is occurring on mini-batches which are randomly sampled at each iteration. However, the numbers that you listed seem lower...