Super Fast and Accurate 3D Object Detection based on LiDAR Fast training, Fast i
SLAM/code 2020. 8. 25. 15:27Super Fast and Accurate 3D Object Detection based on LiDAR
Fast training, Fast inference
An Anchor-free approach
No Non-Max-Suppression
Model:
ResNet-based Keypoint Feature Pyramid Network (KFPN)
Inputs: Bird-eye-view (BEV) maps that are encoded by height, intensity, and density of 3D LiDAR point clouds.
Outputs: 7 degrees of freedom (7-DOF) of objects: (cx, cy, cz, l, w, h, θ)
cx, cy, cz: The 3D center objects.
l, w, h: length, width, and height of the bounding box.
θ: The heading angle in radians of the bounding box.
Objects: Cars, Pedestrians, Cyclists.
The pre-trained model has been released in the repo.
Source code: https://github.com/maudzung/Super-Fast-Accurate-3D-Object-Detection
