Pierre Moulon

Feel free to perform a PR with the change you suggest. Using tricks to make matching faster is always having a cost, and so we are looking forward to feedback...

Thank you for this PR. I will perform a review on it ASAP. Do you have any timing information to provide?

Thank you. I think that a complete benchmark to compare the actual SIFT/SIFT anatomy/GPU Sift would be valuable later on. We can also test in a near future the repeatability...

@rperrot This is this SIFT implementation that is used https://github.com/Celebrandil/CudaSift

Thank you @saihv for the benchmark about the timing. I'm also interested into repeatability (capability of being as good as VLFeat or SiftAnatomy). Often GPU implementations are less good that...

Certainly this is due to a difference of scale (input pixel range, or normalization issue). We will need to investigate in order make those threshold match. We see here the...

Thank you @saihv. Often GPU implementation are less accurate, and here we see it (for some scene the results are a bit different). The good things is that curves look...

@hiakru Feel free to make a PR in order to improve the existing repository.

Thank you. I will take a look during the week and let you know. FYI when you run install there is a manifest file that trace all the file that...

Let's do step by step. Starting with the LIB is a good start