AVOD
Aggregate View Object Detection
ONE
Kitti Object Detection Dataset
TWO
AVOD paper
THREE
AVOD code
Kitti Object Detection
Dataset
http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d
Introduction
data collecting platform:
•
•
•
•
2 × PointGray Flea2 grayscale cameras
2 × PointGray Flea2 color cameras
1 × Velodyne HDL-64E rotating 3D laser scanner
1 × OXTS RT3003 inertial and GPS navigation sy
stem
4 × Edmund Optics lenses
•
tasks of interest :
• stereo, optical flow, visual odometry, 3D object dete
ction and 3D tracking.
Ground truth:
• Velodyne laser scanner and a GPS localization syst
em.
3D Object Detection Dataset
class:
•
’Van’, ’Car’, ’Truck’, ’Pedestrian’, ’Person (sitting)’, ’Cyclist’, ’Tram’ ,’Misc’
3D bounding box overlap :
• Car:70%
• pedestrians 、cyclists:50%
Difficulty:
• Easy: Min. bounding box height: 40 Px, Max. occlusion level: Fully visible, Max. truncation: 15 %
• Moderate: Min. bounding box height: 25 Px, Max. occlusion level: Partly occluded, Max. truncation: 30 %
• Hard: Min. bounding box height: 25 Px, Max. occlusion level: Difficult to see, Max. truncation: 50 %
Point Cloud
• N*4 matrics:
(x,y,z, reflectivity)
• Project to image coordinate:
x = P2 * R0_rect * Tr_velo_to_cam * y
Evaluation
• Average Precision(AP):
• Average Orientation Similarity (AOS) :