Team Bagged Data Hosting (Watch This Space)

[AlexanderNorris]

Hi All,

I spoke to all the 2018 teams (and Louisiana) while we were in Hawaii and discussed the possibility of sharing our bagged data from our WAM-Vs.

Our main goal is to provide a comprehensive image set for future development of computer vision solutions, which may involve a coordinated labeling effort once the data has been collected. Any other data which can be provided could potentially be used to develop other perception and control systems and would be appreciated as well.

Since we would like to have data from outside Hawaii, we're also looking for any data captured from WAM-Vs during testing although we would prefer it if this testing was done with the types of buoys used in the competition so we can standardize our data. If you've competed before or have older data which could be of use, please let us know and we'll gladly host your data as well.

If your team has been using ROS, we would love to receive your raw bag files.
If your team hasn't been using ROS, it would be best if we could receive just your video/image data to simplify the work on our end.

Hitesh and I are working to set up a hosting platform for all this data and we should be able to get something sorted out shortly.
In the meantime, it would be fantastic if teams could get their bagged data ready to upload.

Watch this space for updates and I would also like to send emails as a backup communications channel. I currently have at least 1 email address for each team from RobotX 2018 apart from Georgia Tech and Louisiana Lafayette. I would appreciate it if these teams could give me an email address which we can use for this.

[brennan]

Thanks alex, this is a great idea.

I'm not sure what you have in mind as far as post-processing the bag files (if any), but particularly for the teams not using ROS, it'll definitely be useful to have the images extracted from all of the bag files (as an example our team pulls images from the compressed stream at 1s intervals so we don't end up with a extremely large set of similar images). This is also because all image classification neural network implementations that I am aware of require that all images are of identical resolution. So the camera variance (which at a minimum, can vary with aspect ratio and resolution) between the teams will require that we post-process all of the images anyway. This could be accomplished in real-time with some kind of cropping+scaling package, but I think for simplicity sake the majority of us will end up recreating the full data set as most appropriate for our particular system manually. So not having to jump through the hoop of extracting the images from the bag file will be useful for most of us.

This is not to say that uploading the bag files is not useful--I would still love to have access to those for reasons outside of machine learning. If any of the more-accomplished teams are willing to share their raw bag files, that would be a huge learning experience for our team for sure. We are of course willing to share our bag files, but if any learning is derived from it, it will probably be "here's an example of what not to do"...

[Vinny]

Hey Alex,

Vinny from Georgia Tech here. Add me to the email list: vruia3 [a.t.] gatech DOT edu

[Doc_Vaughan]

I'm the best contact here at UL Lafayette - [email protected]

[michaelbenson]

Hey Alex,

You can feel free to add my email to the list ([email protected]), as I'm a bit removed from FAU's campus, and have most of the Owltonomous data as well as Travis.

I think an important part of this is going to be determining the scope of the data you'd like to collect, and how it's curated. Even if every team used ROS at RobotX this past December, not every teams' bag files will be made equal. This isn't meant to be a knock at anyone, I know personally have developed a number of packages for ROS only to find out a month later it's already been done and is part of the official guidelines (see: my 4 abandoned repos for lidar messages and processing). Different topic names, reference frames (and frame names), and custom message types will plague anyone that tries to view another team's data to some extent. It might be helpful to set up a guideline for uploaded bag files - perhaps only some specific information, such as camera data, lidar data, GPS fix, etc. - and appropriate topic names, reference frames, and message types. I think it will be far easier to have each team curate their data accordingly before they upload it than it would be for one person to try to figure out what the heck /pwd_points_ned_v2 from another school's data set is.

With all that being said, I think it's an excellent idea to try to share data, both for old and new teams. More than a few teams seem to lack the ability to test their WAM-V sufficiently before the competition, so having real data to validate their software systems before competition would go a long way to improve the quality of competition, especially given that simulation isn't sufficient to thoroughly vet your systems (as we learned the hard way this year).

[brian.bingham]

Alex,

This is a wonderful effort. Please include me in the comms [email protected]

All the best,
Brian