bernwang / latte

LATTE: Accelerating LiDAR Point Cloud Annotation via Sensor Fusion, One-Click Annotation, and Tracking
Apache License 2.0
398 stars 157 forks source link

Is this possible to add a functionality for 3-D bounding box annotation? #2

Closed zhou-git closed 5 years ago

zhou-git commented 5 years ago

Dear Bernwang,

The tool works flawless, Thank you very much for your work! I am wondering if it is possible to add a functionality to do 3-D bounding box annotation? Specifically, the ability to specify the ground level and the height of the object?

bernwang commented 5 years ago

Hi zhou-git,

I apologize for the late response. There has been a lot of work on the annotator that I am excited to push. One of the features is to add 3D bounding box annotation. I am thinking of automatically extracting the height of the bounding box after drawing a 2D bounding box. Do you think you would prefer that over manually specifying the height of the object?

Thanks, Bernie

zhou-git commented 5 years ago

Hi Bernie,

It is great that you still works on this project! Automatically extracting object height seems interesting but I not sure about the precision, which is very important to manual annotations. A typical case is like annotating a car with no points available on the wheels, then it becomes tricky to determine the ground plane of that 3D box. I kind of thinking that a easy to use interface is most important for a manual annotation tool in the first place. But of cause it is a open-source project, so it may be too much to ask.

Best, Jin

songanz commented 5 years ago

Hi Bernie,

Thank you very much for your talented original work. I found that the most helpful open source project. I am also interested in having that 3D bounding box annotation feature and I think the automatic annotation bounding box would be enough for now. If we can manually adjust it would be better, but I really understand that you are pretty occupied by your own work and you have very litter time for that.

Best, Songan Zhang

songanz commented 5 years ago

Hi Bernie,

It is great that you still works on this project! Automatically extracting object height seems interesting but I not sure about the precision, which is very important to manual annotations. A typical case is like annotating a car with no points available on the wheels, then it becomes tricky to determine the ground plane of that 3D box. I kind of thinking that a easy to use interface is most important for a manual annotation tool in the first place. But of cause it is a open-source project, so it may be too much to ask.

Best, Jin

I have developed myself. If you want to use it, please find it at https://github.com/songanz/LiDAR-annotator

davodogster commented 3 years ago

Hi Bernie,

It is great that you still works on this project! Automatically extracting object height seems interesting but I not sure about the precision, which is very important to manual annotations. A typical case is like annotating a car with no points available on the wheels, then it becomes tricky to determine the ground plane of that 3D box. I kind of thinking that a easy to use interface is most important for a manual annotation tool in the first place. But of cause it is a open-source project, so it may be too much to ask.

Best, Jin

Hi Jin @zhou-git but the ground plane will just be the lowest Z point possible because the car wheels always touch the ground?