openMVG icon indicating copy to clipboard operation
openMVG copied to clipboard

OpenMVG 2019 Survey

Open pmoulon opened this issue 6 years ago • 7 comments

We want to listen to the community, better understand your usage of the library and react accordingly. In order to do so, we created a survey and we are really eager to see the results.

Please fill out this survey https://forms.gle/7MHuApGCgh7aDdgQ7.

We will analyze results, share them next year and will use them for the decision-making process in the future.

Please re-share this to your friends and remember that we are always looking for new contributors and users!

Happy holidays :christmas_tree: :santa: :mrs_claus: :snowman: :christmas_tree:

pmoulon avatar Dec 24 '19 16:12 pmoulon

Just a friendly reminder to some OpenMVG contributors: @hcjghr @cbalint13 @mitjap @bjornpiltz @bchretien @rhiestan @NikolausDemmel @KevinCain @sebastienchappuis @rfabbri @mdaiter @Joeppie @Meekohi @john-davies @TheDoubleB @orva @lcalvet And some active fork: @daisatojp @meder411 @buaalinpg @belosthomas @wenyinwei @frankie-yanfeng @lijgame @oppenfuture @shauidu @rockokw @qiwenwang @Pandinosaurus @humenbergerm @roby23 @kant @kennege @romi @gaoshuai @shuhilya @smithangshu @zhangchenghgd @tdahlke @Mirmix @eivan @larshaalck @donlk @tkonolige @belavina @Ella2le

pmoulon avatar Dec 26 '19 16:12 pmoulon

Hi, Pierre. submitted it! Thanks for your work again. happy holidays.

gaoshuai avatar Dec 27 '19 05:12 gaoshuai

Thx you @gaoshuai!

pmoulon avatar Dec 27 '19 07:12 pmoulon

@florentPoux Feel free to help by taking the survey ;-)

pmoulon avatar Dec 30 '19 23:12 pmoulon

Thank you, everyone, for your contribution. The results are very informative and we are hoping to run a similar survey every year.

  • OpenMVG users

    • We are happy to see that we have a long tail of regular users
      • 40% of the users are using OpenMVG for more than 3 years
    • Mobile phone camera and ground capture are the dominant way to provide input image for your OpenMVG projects
    • 44% of the users use 1000+ images in their projects
  • OpenMVG SfM pipeline

    • Incremental SfM remains the "to-go" pipeline
    • 30% of our community is building pipelines on top of OpenMVG
  • OpenMVG code quality

    • the community recognizes that OpenMVG has provided a lot of effort to bring SOTA (State Of The Art) algorithms to everyone
  • Most desired feature by the community

    • SLAM, Laser + SfM , Image Matching speed, MVS, Accuracy, SfM speed, Feature Detection & Matching, Geodesy are the most desired area for new features (Come and join us!).

Here is the detailed result of the survey:

How the community is linked to OpenMVG:
1 new - 5 expert image6
image12
image20
How the community is using OpenMVG:
image19
image2
image13
Strengths and area of improvement of OpenMVG
image7
image18
image11 By order (SLAM 42%, Laser + SfM 38%, Image Matching speed 36%, MVS 36%, Accuracy 36%, SfM speed 34%, Feature Detection & Matching 34%, Geodesy 32%, More Sfm Pipeline 24%,... )
Which kind of cameras & images the community is using with OpenMVG:
image3
image16
image5
image14
Knowledge about the lib feature and license:
image4
image9
Who are the OpenMVG users:
image1
image17
image15
Does the community is ready to contribute:
image10
Random comments from the community
Thank you very much. I do not take open MVG for granted
Thanks for your work!
OpenMVG is a fantastic opensource framework
This library is incredible, thanks for your great work!
I will do my best to keep supporting OpenMVG !

pmoulon avatar Jan 20 '20 22:01 pmoulon

Thanks Pierre, awesome to see some stats on how this project is being used! Do you have any thoughts as the author on where contributions are especially needed?

Meekohi avatar Jan 21 '20 14:01 Meekohi

Hello @Meekohi , I really appreciate that you are asking the question back! It is nice to see that the community also cares about the vision of the OpenMVG developers.

Here is my actual view or vision on this:

Awareness

  • I still see a need to make Multiple View Geometry easily accessible to extend awareness of the power of photogrammetry.
  • OpenMVG is used by multiple laboratories and industry, either for training in photogrammetry or to run commercial software. Making those entities speaking about their usage of OpenMVG to the community could be beneficial for the OpenSource eco-system.

Accessibility

  • Despite OpenMVG is "easy" to use, writing an algorithm once and be able to use it on any platform is a challenge, especially with the embedded platform and the community of C++ developer that is becoming smaller. Today, running a reconstruction is as simple as running a docker and calling a command line. I would have never imagined 7 years ago that I would make an entire pipeline accessible to anyone in two lines of code (Docker init, Docker launch). Since today most of the developers are leveraging Python, having people to help to develop a Python wrapper on top of OpenMVG would be beneficial to the project and the community.

Community

  • OpenMVG is having an active community. Having a community is important since you can foster new development depending on the need of the users. Our community understands that the OpenSource contributions are mainly made on free time and so having an accurate implementation of a state of the art algorithm is better than having nothing to rely on. The user feedback about our code is impressive and we want to continue to have this dialog.

Speed

  • OpenMVG focuses on accurate implementation, so sometimes the community complains that the pipeline is not fast enough. Implementing speed in an algorithm is taking a lot of time and companies all around the world are paying millions of scientists and engineers to work on optimization rather than easy to read and modify implementation. OpenMVG developers are doing their best to provide new implementation and integrate them with the possibility to switch between methods (see recent development made towards multiple methods for 2-view triangulation, or the Kneip, Ke and Nordberg p3P solver).

Relevance

  • OpenMVG has focused first on implementing all the tools that SfM, Multiple View Geometry requires (features, matching, tracking, multiview, robust estimation) and then implemented pipeline on top of that using a common backbone (the sfm_data container and the sfm_data provider). This API granularity enables anyone to use either algorithm, SfM pipeline very easily and enables any developer to bring their own customization very easily. For sure there is some flaws and inconsistency in naming due to the long history of the project, but what the community cares about algorithms and pipeline is there and it is what matters the most.
  • Bringing knowledge and accurate implementation is the main objective of OpenMVG. ->

OpenMVG credo is: "Keep it simple, keep it maintainable". OpenMVG targets readable code that is easy to use and modify by the community.

  • OpenMVG is full of "hidden' header only gems (unordered feature tracking, SVG drawing, Camera models, ColorHeatMat, image processing, matching in N dimension(brute force and approximate)... ). I personally use daily those tools daily at work and it helps me to a powerful Scientist!

  • Machine Learning and Deep Learning in bringing more and more tools for describing images or performing dense image alignment. OpenMVG would be happy to welcome contributions that would allow testing those new methods in SfM pipelines.

What to expect for 2020

  • OpenMVG will candidate to Google Summer of Code and expect to have support to find contributors (the third time we are trying). But sadly, GSoC selects multiple large project many years in a row and let few places for project that are looking for help to grow.
  • OpenMVG developers are working towards enhanced scalability of its CBIR pipeline (image matching). Pre-emptive matching and a nice implementation of VLAD (Vector of Locally Aggregated Descriptors) is nearly ready to be released.
  • OpenMVG 2.0. We understand that some naming convention of OpenMVG is disturbing, we are planning to fix that when we are gonna release OpenMVG 2.0 since a new major revision allows the team to break the API.
  • We also hope to release some new SfM pipelines and new minimal solver implementation
  • Dreaming: Why not believe we could hit 3K Github Stars and 2K Forks in 2020!

Looking for contributors and scientific contributions

  • The CV, SFM, MVS community is active and sometimes reluctant towards sharing code. OpenMVG is offering a platform for every student, scientist to anyone to publish and demonstrate the superiority of their implementation to the community. We would love to encourage the researcher and the industry to contribute back to OpenMVG and not just using it.

Acknowledgement

  • I would give a huge KUDOS to @rperrot & @rjanvier for their active contributions & help towards user's issues. Without them, OpenMVG 1.5 would not be as complete as it is today! It is rare to have developers that are also using OpenMVG on large real projects for their real jobs.
  • I would give credits to @cogitas3d, @lucaarcteam, @alearcteam, @sflotron, @mitjap and all the contributors I'm not citing here.

pmoulon avatar Jan 24 '20 22:01 pmoulon