Skip to content

Gpu implementation #48

Open
Open

Description

@edgarriba

Hi,

What do you think about using the GPU during the matching process or BA? That could a nice improvement in terms of speed.

Activity

  1. paulinus commented on Oct 30, 2015

    @paulinus
    Contributor

    Best fit for GPU is feature extraction and matching. For this, calling opencv's implementation is what would make more sense. There is no python API for that yet AFAIK.

    BA is probably too ambitious

  2. edgarriba commented on Nov 3, 2015

    @edgarriba
    ContributorAuthor

    Last week I received a couple of GPUs, I'll try to make a Python wrapper to use it in csfm.c

    By now, in cudafeatures2d are only available ORB and BruteForce matching.
    http://docs.opencv.org/master/d6/d1d/group__cudafeatures2d.html#gsc.tab=0

  3. gateway commented on Dec 6, 2016

    @gateway

    I wanted to see if their was any updates on adding gpu support to this? See other discussion over at Open Drone Map to figure out if processing can be accelerated with GPU/s . OpenDroneMap/ODM#92 (comment)

  4. theomessin commented on Apr 2, 2019

    @theomessin

    Any sort of update on this?

  5. 5y commented on Jul 5, 2019

    @5y

    Hello, Is there any GPU feature yet ?

  6. YanNoun commented on Mar 15, 2021

    @YanNoun
    Contributor

    A tentative GPU branch is pending for merge here : #623

    Wasn't yet merged, as the speed-up wasn't dramatic compared to multithreaded CPU. More detailed comparison on > 20K features might prove it is maybe faster, but for now, multi-threaded CPU still wins.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions