Mårten Björkman

It's good to hear that you got it running. As soon as I have time, I will try to explore Turing and see what CUDA 10 has to offer. Most...

I managed to run it on a Jetson TK1 before. It could be that you don't use CUDA 9.0 or later, since __any_sync used to be called __any in earlier...

You mean [Yu & Morel, 2011], not [Mikolajczyk & Schmid, 2002], right? Do you have a feeling for how many angles you need to test in practice?

It's really a brute force method and I'm not sure it's the most effective one. The DoG responses might be more invariant to rotations than the descriptors, which means that...

I might attempt a GPU implementation, together with Alessandro Pieropan, but that depends on whether we'll find time for it. Most things are relatively easy to parallelize, even if there...

To speed up the CPU version, you should try to focus on the memory footprint and access patterns. For example, instead of first creating the pyramid and then detecting features,...

On a Tesla K40c I can compute the scale space in about 6.7 ms, excluding GPU-to-CPU transfers. The sequential nature of the computations makes it hard to limit the number...

I am referring to the grid size h in Weickert et al [21] at equation (8). For each increasing octave, h is doubled in terms of the original image. You...

By the way, subsampling is done with cv::resize and INTER_AREA for interpolation, which means that, unlike what seems to be assumed in the rest of the code, the scale down...