extract 2d-3d bounding box corners

[sirine90]

Thank you for sharing your code.

I want to draw the projected ground truth 3D bounding boxes onto the 2d image, however, I couldn't extract the corners of the 3d boxes in 2d correctly. Do you have an idea how can I do it?

Thanks!

[HaolinLiu97]

Hi, I have a small code snippet about projecting bbox corner into 2D image. The obj_matrix can refer to the tran_matrix, in the prepare data, you can try if it is able to draw correctly.

def get_bbox_corners(bbox_size,obj_matrix,wrd2cam_matrix):
'''
get bbox in camera coordinate
'''
s_x, s_y, s_z = bbox_size[0], bbox_size[1], bbox_size[2]
verts_cannonical = [[- s_x / 2, - s_y / 2, - s_z / 2],
[- s_x / 2, - s_y / 2, s_z / 2],
[- s_x / 2, s_y / 2, - s_z / 2],
[- s_x / 2, s_y / 2, s_z / 2],
[s_x / 2, - s_y / 2, - s_z / 2],
[s_x / 2, - s_y / 2, s_z / 2],
[s_x / 2, s_y / 2, - s_z / 2],
[s_x / 2, s_y / 2, s_z / 2]]
verts_cannonical=np.array(verts_cannonical)
verts = np.dot(obj_matrix[0:3,0:3], verts_cannonical.T).T+obj_matrix[0:3,3]
verts=np.dot(wrd2cam_matrix[0:3,0:3],verts.T).T+wrd2cam_matrix[0:3,3]
verts[:,0:2]=-verts[:,0:2]
return verts

def project_points2img(points,K):
'''
:param points: n3, in camera coordinate
:param K: 33 intrinsic matrix
:return:
'''
points=np.dot(points,K[0:3,0:3].T)
x_coor=points[:,0]/points[:,2]
y_coor=points[:,1]/points[:,2]
z=points[:,2]
img_coor=np.concatenate([x_coor[:,np.newaxis],y_coor[:,np.newaxis],z[:,np.newaxis]],axis=1)
return img_coor

[HaolinLiu97]

I think it maybe the problem that the image in the prepare data is already downsampled by 2. You can try to draw with half the values of the image coordinates.

[sirine90]

It fixed the problem. Thank you very much! :)