PointCloudLibrary Survey 2020

Hi ROS users,

PCL is a common library used for processing Point Clouds in ROS. As such, we’d like to get an insight into how you use and acquire the library, what features are most anticipated as well as what in PCL is important to you. We’ve created a survey that allows us to understand some of these questions or helps us ask better questions.

If you or your team uses (or used) PCL (directly or via perception_pcl), please find the survey at https://tinyurl.com/pcl-2020-survey. If you have shifted (or plan to shift) to some other solution, please inform us the reasons for that at the same link.

It’s a small survey of 10 questions. We’ll keep it open till 2020-Aug-20 and will provide a summary in this thread by Aug 30th.



If you’re able to edit, it might be good to clarify on the 1-N scoring questions that 1 or N is the highest score. I assume it was N but wanted to clarify.

Keep up the awesome work and I’d love to chat more or learn more about the PCL roadmap and wishlists. It’s been awesome seeing the recent documentation push!

I’ve updated the questions to show that #1 is highest rank. Sorry for the confusion.

Thanks for your kind words. BTW, stay tuned for some good news in the wish-list area if all continues to go well in this GSoC.

1 Like

… maybe you should invert my scores then :sweat_smile:

I wanted todo use PCL in my tool Darknet ROS 3D on ROS2. I used it in ROS1 but in ROS2 I had to use tf2 for point cloud frame transformation

Anybody who fills this out, please check out the new PCL website first https://pointclouds.org/ It’s lovely!

@fgonzalezr1998 I’m sure the people over at perception_ros are working hard to get it ready.

@simonschmeisser Thanks for the kind words. The website is also open source: https://github.com/PointCloudLibrary/PointCloudLibrary.github.io, so feel free to contribute to the website as well.

We also have a new home for tutorials: https://pcl.readthedocs.io

Both of these are work in progress, but I hope they are more accessible to the community to edit, search and update than the previous iterations.

Just a reminder: closing the survey on 20th (about 1200 UTC) of August

I’m cutting it close to my self-imposed deadline for the results. Please bear in mind that this is a survey within a ecosystem and is not a representative sample of the kind of users that PCL has.

  1. How long have you used PCL?

    • 2 or more years: 42%
    • 6 months to 2 years: 35%
    • Less than 6 months: 25%
    • Never: 1 kind soul
  2. Preferred OS for using PCL (percentages don’t add to 100% for obvious reasons)

    • Debian and friends: 85% (wasn’t a big surprise that this was the largest group, but the margin was shocking)
    • Windows: 20%
    • MacOS, ArchLinux, BusyBox, Gentoo, NixOS: 1 representation each
  3. Best qualities of PCL:

    1. Feature-set
    2. Popularity
    3. Ease-of-use, tooling, speed
    4. Design
  4. Most common modules:

    • filter (65%)
    • segmentation (22%)
    • common, io, kdtree, registration, visualization (15-17%)
    • features, geometry, keypoint, octree, tools, surface (9-12%)
  5. Primary way to get PCL:

    1. apt (67.5%)
    2. GitHub repository (13%)
    3. VcPkg/GitHub release/Homebrew: single digits
    4. aur, conda, docker: 1 representation each
  6. What tools are used for working with PCL:

    • MSVC (16%) is lower than user count at 20%
    • 54% respondents use the newer catkin/ament/colcon
    • 43% still prefer the old rosbuild/catkin_make method
    • GCC is heavily preferred compared to clang
  7. What should PCL focus on improving:

    1. Tutorials/guides (clear winner)
    2. Usability, Existing algorithms
    3. Speed
    4. API design (far behind)
    5. Others (not insignificant, but far behind)
  8. What new features should PCL bring:

    1. Better GPU (81.5%)
    2. Python bindings (63%)
    3. Faster releases (13%)
    4. Others with a single representation:
      • Better integration with Deep Learning frameworks
      • Bench-marking functionality to help researchers
      • Bindings for Rust, Javascript
      • RGBD/LiDAR algorithms that work out-of-the-box
      • Better multi-threaded algorithm support
  9. Can we contact you?

    • Yes: 45% (that’s a good turnout, don’t you think? :smile:)
  10. Other comments:

    • Up-to-date installation is hard using package managers
    • Large point clouds aren’t supported
    • Thanks :blush:

Key takeaways:

  • The popular modules line up with the issues/questions on PCL GitHub and discord server
  • Windows users responses don’t match the issues/questions count. Possible reasons:
    • vocal minority
    • no interest in responding to surveys
    • lack of people using ROS on windows (see Q6)
    • lack of rigor in PCL for testing on Windows, and many more
  • Possibly release multiple versions of PCL on the apt repo to allow people to use the latest versions.
  • This comes as a relief, but the topics chosen for GSoC are what the respondents want. GSoC-2020@PCL