alexstaravoitau / kitti-dataset Goto Github PK

Hi,

Thanks for sharing!

The visualized 3D bbx shape looks unnatural to me. I find that your way of defining local coordinates and converting the yaw angle to its corresponding rotation matrix is different from kitti dev kit:
trackletBox = np.array([
[-l / 2, -l / 2, l / 2, l / 2, -l / 2, -l / 2, l / 2, l / 2],
[w / 2, -w / 2, -w / 2, w / 2, w / 2, -w / 2, -w / 2, w / 2],
[0.0, 0.0, 0.0, 0.0, h, h, h, h]
]),
rotMat = np.array([
[np.cos(yaw), -np.sin(yaw), 0.0],
[np.sin(yaw), np.cos(yaw), 0.0],
[0.0, 0.0, 1.0]
])

as opposed to the kitti dev kit (Matlab):
% 3D bounding box corners
x_corners = [l/2, l/2, -l/2, -l/2, l/2, l/2, -l/2, -l/2];
y_corners = [0,0,0,0,-h,-h,-h,-h];
z_corners = [w/2, -w/2, -w/2, w/2, w/2, -w/2, -w/2, w/2];

% compute rotational matrix around yaw axis
R = [+cos(object.ry), 0, +sin(object.ry);
0, 1, 0;
-sin(object.ry), 0, +cos(object.ry)];

Could you explain a bit why they are different?

Thanks

Hello, i really enjoyed reading your article and your repository about the kitti dataset exploration.
Im an computer science student and would really like to use your code for further kitti dataset exploration and machine learning tasks.
Would be great if you could add a licence to your repository so i can refer to it.
;)

Hi,

Could you please teel me How to extract distance from camera to object?

Thank you.

Hi Alex,
I am trying to execute your code & use some parts of it to improve sensing for self driving cars. I am constantly facing null error in Xml file and there's a syntax error in accessing same localhost file. Please go through the drive link for reference. I am fairly new to this and your clarification might help me for my graduate project.
Awaiting your reply. Thank you for your valuable time.
Kind Regards,
Chinmay Upadhye

https://drive.google.com/open?id=1I12Ge_SEFK8XSbn760Xp_2UjmlN6ULMc

add import itertools in the file

Hi, can you help me with a question?
I've been working with Velodyne for a while, and kitti is a very good dataset to me because I don't have to label stuff myself. But the problem is that in my project I use raw points from Velodyne, which only has distance and reflectivity and comes in firing order. Velodyne data from kitti is not stored in firing order, nor is it possible to convert (x, y, z) coordinate back to raw point. Actually the number of points in each file is not even multiple of 64. So could you come up with any idea how I could order the points in firing order? Or at least I could put all 64 points with same rotation together?
Many thanks.

Hi
Where did you obtain the labels from?

Hi, I hope to extract the sub-cloud of the object within its gt 3D bounding box as inputs for my model. Are there any efficient methods I could use to do that?

I followed the step and met this problem. Can you help me find where I am wrong? I think I am wrong at this 'return'. But I tried a lot of method but they didn't work. Thx!

Hi,
Thanks for this awesome project. Really helped me a lot.

I had a question, rather an issue that I am not able to solve.

I used groundtruth ktti poses and converted it to tum (timestamp x y z q_x q_y q_z q_w)format using the code here: https://github.com/MichaelGrupp/evo.
Further, I used cartogrpaher to run SLAM on a sequence 005(2011_09_30 - 0018). I have to calcualte the error between both the poses for my research.

However when I plot both the pose maps, they have different orientation, although the overall structure is the same.
Here is the glimpse of the same.

I am not able to align it, which is required for finding ate/rpe between the poses.

The trajectory txt required to plot this image is as followd:
kitti05.txt
&
kitti_2011_09_30_drive_0018_synced.bag.txt

It woulbe great if someone could throw light on this and help me figure out the correct way to aligb both the trajectories.

Hey is there a script available to convert the KITTI Dataset labels into COCO Datset label format?
I have found the script for the other way around. But I am looking for KITTI to COCO conversion.

Thanks!

Hi,
Thank you for your work. I have problem when running your code:

  File "test.py", line 75, in load_tracklets_for_frames
    for translation, rotation, state, occlusion, truncation, amtOcclusion, amtBorders, absoluteFrameNumber in tracklet:
TypeError: iter() returned non-iterator of type 'list'

Please help me to fix this.

Best regards