Development potential in Localization, Computer Vision and Deep Learning

I’m eyeing contributing a ROS2 package for a project in my master’s programme in signal and image processing. Since ROS packages don’t have a central official repo like OpenCV for example, I find it hard to get an overview of what kind of algorithms are well supported by existing packages and where there are gaps I could look at filling.

So I am asking you: Have you come across a topic or algorithm where you were left wanting by the existing implementations or where there are none yet?

Since ROS packages don’t have a central official repo like OpenCV for example, I find it hard to get an overview of what kind of algorithms are well supported by existing packages and where there are gaps I could look at filling.

You mean like this https://index.ros.org/?

So I am asking you: Have you come across a topic or algorithm where you were left wanting by the existing implementations or where there are none yet?

Personally, I think it might be better to pay down some technical debt before trying to create new features out of whole cloth. Not only will this have a bigger impact on the project, but it will allow you to familiarize yourself with the existing ecosystem. Is it sexy research work? No, but it is really important to the health of the project. I would argue that in terms of career it is almost as important to maintain a widely used project than produce a one-and-done research package that lasts for a distro or two. I would also suggest you take a look at REP-2004 so that you have reasonable expectations around what a “good” package looks like in terms of code quality.

As to where the gaps are, I’ve been saying we need a “NERF to point cloud to Gazebo simulation environment” pipeline / package for quite some time now. If you make it happen I will absolutely slip a letter full of ROS swag in the mail.

2 Likes

You mean like this https://index.ros.org/

Oh shoot! That helps, thanks!

And yea I totally get your point about what is most valuable to the project. I was looking into the suggested first contribution issues as well. Unfortunately, the only way I can dedicate the needed time is through a uni project, which ought to be centered on the maths and algorithms rather than software engineering.

Your NeRF suggestion is an interesting one! I’ll read into it some more.