search

kujason / avod

Code for 3D object detection for autonomous driving
MIT License
931 stars 349 forks source link

Planes #86

Closed radler12 closed 4 years ago

radler12 commented 6 years ago

Can you explain what contais planes folder in your framework and how you generate these files?

Best regards,

walzimmer commented 6 years ago

The planes folder in the KITTI dataset contains the ground planes of the ego vehicle. For each camera image (frame) you have a corresponding ground plane that was calculated from the IMU data (roll, pitch, yaw) of your vehicle. The format is the following:

Width 4 Height 1 -7.051739e-03 -9.997791e-01 -1.980150e-02 1.680367e+00

The vector [-7.051739e-03 -9.997791e-01 -1.980150e-02] defines the normal vector of the plane. It is almost [0 -1 0] which means that the plane is nearly even. The forth value 1.68 is the height of the camera relative to the plane.

The ground planes are used to calculate the 3d anchor grid.

villanuevab commented 6 years ago

What are the Width and Height values used for?

Width 4
Height 1

Thanks for your time!

YunzeMan commented 6 years ago

@villanuevab This is the dimension of the plane matrix., which is 4 x 1 like [-7.051739e-03, -9.997791e-01, -1.980150e-02, 1.680367e+00].

FrankCAN commented 6 years ago

@walzimmer For the detection training data, where did you find the IMU data for ground plane generation? And for the vector [-7.051739e-03 -9.997791e-01 -1.980150e-02 1.680367e+00], I think the fourth parameter is 1.65m, but how did you calculate your fourth parameter which is 1.680367e+00, and why doesn't it be a constant value?

Many thanks for your help, very much appreciate for that.

lucasjinreal commented 5 years ago

Does anyone knows how to using a constance ground plane for prediction a video sequence data?

mengmengQi commented 5 years ago

The planes folder in the KITTI dataset contains the ground planes of the ego vehicle. For each camera image (frame) you have a corresponding ground plane that was calculated from the IMU data (roll, pitch, yaw) of your vehicle. The format is the following: Width 4 Height 1 -7.051739e-03 -9.997791e-01 -1.980150e-02 1.680367e+00 The vector [-7.051739e-03 -9.997791e-01 -1.980150e-02] defines the normal vector of the plane. It is almost [0 -1 0] which means that the plane is nearly even. The forth value 1.68 is the height of the camera relative to the plane. The ground planes are used to calculate the 3d anchor grid.

Hi, in fact the fourth parameter change all the time, perhaps it isn't the height of camera. Have you resolved the problem?

staillyd commented 4 years ago

The planes folder in the KITTI dataset contains the ground planes of the ego vehicle. For each camera image (frame) you have a corresponding ground plane that was calculated from the IMU data (roll, pitch, yaw) of your vehicle. The format is the following:

Width 4 Height 1 -7.051739e-03 -9.997791e-01 -1.980150e-02 1.680367e+00

The vector [-7.051739e-03 -9.997791e-01 -1.980150e-02] defines the normal vector of the plane. It is almost [0 -1 0] which means that the plane is nearly even. The forth value 1.68 is the height of the camera relative to the plane.

The ground planes are used to calculate the 3d anchor grid.

@walzimmer Assuming that the unit normal vector of the plane is in the format [x, y, z]. It seems not the coordinate system in which the IMU is located. would you tell me what the coordinate system of plane looks like?

Best regards.