Regarding processing speed

Shiaoming / ALIKED

ALIKED: A Lighter Keypoint and Descriptor Extraction Network via Deformable Transformation

https://arxiv.org/pdf/2304.03608.pdf

BSD 3-Clause "New" or "Revised" License

190 stars 16 forks source link

Regarding processing speed #8

Closed uekoh closed 11 months ago

uekoh commented 1 year ago

Hi. I have compared the processing speed between SuperPoint and ALIKED and SuperPoint is faster. I have the same input images and number of detection points. I used your ALIKED-N(16) model. What conditions are you measuring in your paper?

Shiaoming commented 1 year ago

NVIDIA RTX 2060. I did find that CUDA_NUM_THREADS in get_patches_cuda on different platforms can lead to differences in runtime. In this paper, this parameter was tuned based on my tests using an RTX 2060.

uekoh commented 1 year ago

Thank you for your response. Do you have any suggestions on how to tune it? Do I have to raise or lower CUDA_NUM_THREADS and see what the trend is? In case you are wondering, I am using RTX3060Ti.

Shiaoming commented 1 year ago

I don't have any more information for you to refer to regarding this topic. Mabey you could search for more info about cuda programming~