Method Details


Details for method '3D-GCK'

 

Method overview

name 3D-GCK
challenge 3D vehicle detection
details 3D-GCK is based on the standard SSD 2D object detection framework and lifts the 2D detections to 3D space by predicting additional regression and classification parameters. Hence, the runtime is kept close to pure 2D object detection. The additional parameters are transformed to 3D bounding box keypoints within the network under geometric constraints. 3D-GCK features a full 3D description including all three angles of rotation without supervision by any labeled ground truth data for the object's orientation, as it focuses on certain keypoints within the image plane.
publication Single-Shot 3D Detection of Vehicles from Monocular RGB Images via Geometry Constrained Keypoints in Real-Time
Nils Gählert, Jun-Jun Wan, Nicolas Jourdan, Jan Finkbeiner, Uwe Franke, and Joachim Denzler
IV 2020
https://arxiv.org/abs/2006.13084
project page / code
used Cityscapes data 3D bounding boxes
used external data ImageNet
runtime 0.04 s
NVIDIA V100
subsampling no
submission date January, 2021
previous submissions 1

 

Average results

Metric Value
Detection Score 37.4378
AP 42.5368
Center Distance 96.0878
Size Similarity 70.7106
Orientation Similarity Yaw 81.8904
Orientation Similarity Pitch Roll 99.9731

 

Class results

Class Detection Score
car 67.4544
truck 28.9724
bus 32.2892
train 23.124
motorcycle 32.8675
bicycle 39.9195

 

Results json

This json can be visualized using the tool csPlot3dDetectionResults, which is part of the cityscapesScripts found on Github.

 

Links

Download results as .csv file

Benchmark page