Scripts using for drawing 3d box, can test with only ground truth about 3d-boundingbox HOT 12 OPEN

Thanks

from 3d-boundingbox.

Can you explain the get_calibration_cam_to_image() function in detail? I tried to write it by myself but not work,Thanks

from 3d-boundingbox.

according to jinfagang's repository, 3d_detection, get_calibration_cam_to_image() function is as follows:
def get_calibration_cam_to_image(cab_f):
for line in open(cab_f):
if 'P2:' in line:
cam_to_img = line.strip().split(' ')
cam_to_img = np.asarray([float(number) for number in cam_to_img[1:]])
cam_to_img = np.reshape(cam_to_img, (3, 4))
return cam_to_img

from 3d-boundingbox.

Thanks a lot @yingsen1 I have one more question , How can I get the box3d_dir = '/media/jintain/sg/permanent/datasets/KITTI/tiny_kitti/box_3d/' predi_dir = '/media/jintain/sg/permanent/datasets/KITTI/tiny_kitti/predict_02/'

from 3d-boundingbox.

@yuyijie1995 Hi, you need to predict 3d result first

from 3d-boundingbox.

Thankyou for your reply @jinfagang Is the 3d results from Eval.py 's orient(theta) ,conf , dim ? And the predict_02 is the 2D bbox? I am new to contact 3D detection ..

from 3d-boundingbox.

It seems that something wrong .. bbox's direction is not right

from 3d-boundingbox.

Having the same issue with box direction, @yuyijie1995 did you find out how to fix this?

from 3d-boundingbox.

When I take an image as a test and input the multibin model, I get the orient, dim and conf, and the how to get the R_y from orient, dim and conf?

from 3d-boundingbox.

this method is too old and slow, new end2end method should deploying.

from 3d-boundingbox.

@jinfagang What is the method? can you give me more details? thanks

from 3d-boundingbox.

@yuyijie1995 You can find some progress in here

from 3d-boundingbox.