autoanatation bus or truck not accurate

[suntao1983]

After i deploy the app in ubuntu , i copy my own pcd file into dataset catogory as the example dataset fromat, when i use CTRL+ draw a rectangle enclosing the object, I found there is soming maybe not accurate。

1、draw a rectangle enclosing a car ,the result is approximately right. 2、draw a rectangle enclosing a bus or truck ,the result i see is not right. (see the screen shot)

3、i review the code simplely found the auto annotation bakend process :

1. call the method ‘predict_rotation’ ---use model predict the rotation between the box and the pointcloud target ?
2. use the angle generate the eular_delta ,translate box ,use strategy calculate the the fitted tartget box? Can anyone help me answer my question?

[naurril]

it is caused probably by the incorrect ground removal, as can be seen in the s reenshot there are points under the ground level, and a patch of extra points behind the bus. Is the point cloud combined (inaccurately) from multipe scans?

On Thu, Nov 30, 2023, 10:35 TONY SUN @.***> wrote:

After i deploy the app in ubyntu , i copy my own pcd file into dataset catogory as the example dataset fromat, when i use CTRL+ draw a rectangle enclosing the object, I found there is soming maybe not accurate。

1、draw a rectangle enclosing a car ,the result is approximately right. 2、draw a rectangle enclosing a bus or truck ,the result i see is not right. (see the screen shot) lADPJwKt6DHAnXPNC9DND8A_4032_3024.jpg (view on web) https://github.com/naurril/SUSTechPOINTS/assets/13147226/e6aa4f65-19a7-4735-abd4-efa632a766b1 lADPJw1WZCAWy4DNC9DND8A_4032_3024.jpg (view on web) https://github.com/naurril/SUSTechPOINTS/assets/13147226/f5f29f56-c2bb-465f-8b3c-1dfa263c6c81

3、i review the code simple found the auto annotation bakend process :

1. call the method ‘predict_rotation’ ---use model predict the rotation between the box and the pointcloud target ?
2. use the angle generate the eular_delta ,translate box ,use strategy calculate the the fitted tartget box? Can anyone help me answer my question?

— Reply to this email directly, view it on GitHub https://github.com/naurril/SUSTechPOINTS/issues/179, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABZKCDWXKCSBCVOBQ3ZSJHTYG7WGBAVCNFSM6AAAAABAAORWWCVHI2DSMVQWIX3LMV43ASLTON2WKOZSGAYTONZVHE2DCNQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[suntao1983]

this screenshot comes with the example dataset, if draw a rectangle,the bigger the predict incorrect more happens.