Method

Range-view Retrieval R-CNN [R^2 R-CNN]


Submitted on 16 Feb. 2023 16:17 by
Wang Yihan (NTU)

Running time:0.1 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
First, we encode the range image by novelly proposed feature extractor, then the point cloud is retrieved from the encoded range image. After that, VoxelNet is applied to extract features from 3D points as our 3D backbone. Finally, we employ an RPN and proposed a new pooling method to refine candidate boxes.
Parameters:
None
Latex Bibtex:
None

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Car (Detection) 96.38 % 95.19 % 92.57 %
Car (Orientation) 96.35 % 95.06 % 92.36 %
Car (3D Detection) 90.93 % 82.42 % 77.84 %
Car (Bird's Eye View) 93.02 % 91.17 % 86.62 %
This table as LaTeX

2D object detection results.
This figure as: png eps txt gnuplot



Orientation estimation results.
This figure as: png eps txt gnuplot



3D object detection results.
This figure as: png eps txt gnuplot



Bird's eye view results.
This figure as: png eps txt gnuplot




eXTReMe Tracker