fregu856/3dod_thesis

3D Object Detection for Autonomous Driving in PyTorch, trained on the KITTI dataset.

3dod_thesis's Issues

Source Code

Will the source code be released for this?

GLFW Error: GLX: Failed to create context: GLXBadFBConfig

Hi,

I installed open3d and am getting the below error while running visualize_eval_val.py

$ python3 visualization/visualize_eval_val.py
000006
GLFW Error: GLX: Failed to create context: GLXBadFBConfig
Failed to create window
Segmentation fault (core dumped)

Any solution for this?

Regards
Abhi

is there there instructions to go through only camera detections ?

as in title

camera_trajectory.json

Hi @fregu856,

Thank you very much for your excellent work. It is the most straight-forward code I have ever found.
Just one question, how do you get the camera_trajectory.json file? It seems not working. When I run the visualize_eval_val_seq.py, I have the error:

000000
[ViewControl] ConvertFromPinholeCameraParameters() failed because window height and width do not match.
[ViewControl] ConvertFromPinholeCameraParameters() failed because window height and width do not match.

Frame Rate

What is the frame rate you achieve?

AttributeError: 'open3d.open3d.camera.PinholeCameraTrajectory' object has no attribute 'intrinsic'

Hi,! I met an error when I used Open3D to visualize eval_seq result. And I wanna output the result as a Video. The following message is from terminal.

python visualization/visualize_eval_val_seq.py
PinholeCameraTrajectory read JSON failed: empty trajectory.
000000
Traceback (most recent call last):
File "visualization/visualize_eval_val_seq.py", line 802, in
img_lidar = lidar_img_creator.create_img(gt_bboxes + pred_bboxes + [pcd])
File "visualization/visualize_eval_val_seq.py", line 448, in create_img
vis.run()
File "visualization/visualize_eval_val_seq.py", line 422, in move_forward
ctr.convert_from_pinhole_camera_parameters(self.trajectory.intrinsic, self.trajectory.extrinsic[0])
AttributeError: 'open3d.open3d.camera.PinholeCameraTrajectory' object has no attribute 'intrinsic'

Any solution for this?

Regards,
Cheng

Point level augmentation

@fregu856 thanks for the wonderful code base , just had few queries

can we make the point clouds using augmentation more denser then currently it is
2.can i train the model having only point cloud the rest of all the points eliminated
is there any way we have ground truth removal or estimation to make the model independent of the ground plan
Thanks in advance

NameError: name 'PointCloud' is not defined

Hi @fregu856 ,
I cloned this repo and followed the steps mentioned. But facing an issue while running visualization part: python visualization/visualize_eval_val.py. I have installed open3d also.
error:
$ python visualization/visualize_eval_val.py
000008
Traceback (most recent call last):
File "visualization/visualize_eval_val.py", line 305, in
pcd = PointCloud()
NameError: name 'PointCloud' is not defined

Could you please help me out with this issue?

solved

Hello,

For the camera_trajectory.json, Is this same for all of the frames ?
Can you give me any tips or hint about selecting and control the viewpoint ?

param = ctr.convert_to_pinhole_camera_parameters()
camera_trajectory = PinholeCameraTrajectory()
camera_trajectory.intrinsic = param[0]
camera_trajectory.extrinsic = Matrix4dVector([param[1]])
write_pinhole_camera_trajectory(path,camera_trajectory)

For me now, these code works really bad, every time i need to manually control the view point to get the camera_trajectory file ?

Thanks in advance for your warm help.

how much time you train one epoch?