Convert Cuboid2D to/from KITTI 3D data #1639
Conversation
Signed-off-by: Ilya Trushkin <[email protected]>
Signed-off-by: Ilya Trushkin <[email protected]>
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## develop #1639 +/- ##
===========================================
+ Coverage 81.06% 81.23% +0.16%
===========================================
Files 278 281 +3
Lines 32517 32881 +364
Branches 6607 5289 -1318
===========================================
+ Hits 26360 26710 +350
- Misses 4701 4721 +20
+ Partials 1456 1450 -6
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
| return np.array([a, b, c, d]) | ||
|
|
||
| @staticmethod | ||
| def _get_denorm(Tr_velo_to_cam_homo): |
There was a problem hiding this comment.
Just a question, what is the meaning of Tr_velo_to_cam_homo?
There was a problem hiding this comment.
Calibration matrix Tr_velo_to_cam has a shape of 3 x 4. To project 3D points to the 2D plane, we need to have homogeneous coordinates where each point is represented as a vector with 1 additional dimension.
There was a problem hiding this comment.
This is the projection matrix between Velodyne LiDAR to Camera, where LiDAR contains 4 dimensions ([X, Y, Z, 1]) and Camera contains 3 dimensions ([u, v, 1]). velo stands for Velodyne :)
| 2---3 | ||
| /| /| | ||
| 5-+-8 | | ||
| | 2 + 3 | ||
| 1-+-4 | | ||
| | 5 + 6 | ||
| |/ |/ | ||
| 1---4 | ||
| 8---7 |
There was a problem hiding this comment.
I understood the overall structure, is there any reason to change the bottom and top face?
There was a problem hiding this comment.
I aligned the order of points with Kitti format which describes top face first.
Signed-off-by: Ilya Trushkin <[email protected]>
@sooahleex, documentation is added in 657bf1b. PR description is extended as well. |
Summary
CVS-151427
New features
Cuboid2Dmethods:Cuboid2D.from_3d(dimensions, location, rotation_y, P, Tr_velo_to_cam): Creates a Cuboid2D object from KITTI 3D bbox annotation data. MatrixP(P2in Kitti format context) is a 3x4 projection matrix in the left color camera coordinate system. MatrixTr_velo_to_camis a 3x4 projection matrix between LiDAR and camera coordinate systems.cuboid_2d.to_3d(P_inv): Reconstructs approximate KITTI 3D bbox annotation data (dimensions,locationandrotation_y) from 2D projection coordinates.P_invmatrix is a pseudo-inverse of camera-to-image projection matrix.How to test
See unit test changes
Checklist
License
Feel free to contact the maintainers if that's a concern.