3D Vision

LiDAR-Based 3D Object Detection

Step-by-Step Guide

Welcome to this guide on LiDAR-based 3D Object Detection using the OpenPCDet codebase. In this tutorial, we will explore the OpenPCDet framework in detail and demonstrate how to implement an object detection algorithm across three different levels.

PART 1

Configuring the system with NVIDIA drivers, CUDA, & cuDNN.

PART 2

Setting Up OpenPCDet 3D Open Source via Miniconda

PART 3

Running Demo using Pre-trained weights

PART 4

Training & Testing using Custom Data

Part 3: Running Demo using Pre-Trained Weights

1) Install the Open3D (faster) or mayavi visualization tools

pip install open3d

# or

pip install mayavi

In our case we have installed open3d

2) Download the pre-trained model provided by OpenPCDet

https://github.com/open-mmlab/OpenPCDet/tree/master#model-zoo

I downloaded model-50M for PV-RCNN as 'pv_rcnn_8369.pth' from KITTI 3D Object Detection Baselines

3) go to tools

cd tools

3) Run the demo based on pretrained weights

python demo.py --cfg_file /home/./OpenPCDet/tools/cfgs/kitti_models/pv_rcnn.yaml --ckpt /home/./OpenPCDet/./pv_rcnn_8369.pth --data_path /home/./OpenPCDet/000110.bin