search

utiasSTARS / pykitti

Python tools for working with KITTI data.
MIT License
1.14k stars 239 forks source link

How to read the tracklets and get the box on image? (for both 2d and 3d) #36

Closed lucasjinreal closed 5 years ago

lucasjinreal commented 5 years ago

As title suggested

dakdouky commented 5 years ago

Same problem here! have you found a solution, @jinfagang ?

leeclemnet commented 5 years ago

Currently we only support the raw and odometry datasets. Support for the tracking dataset would be a nice addition if someone in the community wanted to tackle it and submit a PR.

dakdouky commented 5 years ago

I've managed to do it and will make it available soon. thanks :) untitled

leeclemnet commented 5 years ago

Apparently I forgot about https://github.com/utiasSTARS/pykitti/pull/28, which added support for the tracking dataset. I haven't used it personally. Does it do what you need, or have you added something on top of it to get the bounding boxes?

dakdouky commented 5 years ago

In fact, I did not use it. Instead, I build some wrapping on top of Christian Herdtweck XML parser. I'm trying to add a Dynamic/Static annotation to the tracklets.

LukeAI commented 5 years ago

Hey guys, I want to be able to read the raw tracklets and extract the projected 2d bounding boxes and corresponding rgb image. I understand that this is possible with these utils but How can I do this? @MOHAMEDELDAKDOUKY @leeclemnet

DerekGloudemans commented 5 years ago

I implemented bbox plotting for 2d and 3d objects for the tracking dataset. Will port it over to this repo and submit a pull request eventually. For now, the code is in: https://github.com/DerekGloudemans/KITTI-utils/blob/master/track_loader.py

Worth noting that the tracking dataset has very coarse object heights (I don't think they had people manually adjust the heights on this datset and the velodyne data perhaps doesn't contain this info because the objects extend above its range. I added a naive correction with a constant value for each class.