3D Object Detection Using PointPillars

Project Overview

This project implements a 3D object detection pipeline using the PointPillars architecture. It processes LiDAR point clouds and image data to detect objects in 3D space. The model is trained and tested on the KITTI dataset, and the results are visualized using bounding boxes in both 2D and 3D views.

Comparison with Paper Results

Quantitative Comparison Table

Repo	Metric	Overall	Pedestrian	Cyclist	Car
This Repo	3D-BBox	73.3236 62.7845 59.6244	51.4505 47.9423 43.8012	81.8821 63.6617 60.8990	86.6381 76.7495 74.1730
Paper	3D-BBox	78.53 66.42 63.25	55.32 50.11 45.98	85.21 70.35 67.89	90.12 81.45 78.23
This Repo	BEV	77.8532 69.7995 66.6686	59.1663 54.3520 50.5038	84.4268 67.1311 63.7362	89.9664 87.9153 85.7659
Paper	BEV	81.45 72.89 69.75	63.12 58.78 54.32	87.56 74.32 70.89	92.01 88.43 85.23
This Repo	2D-BBox	80.5047 74.5842 71.4837	64.6100 61.3595 57.6168	86.2569 73.0603 70.1746	90.6471 89.3327 86.6598
Paper	2D-BBox	83.21 76.42 73.15	67.54 64.23 60.12	89.98 76.12 72.89	93.45 90.23 88.45
This Repo	AOS	74.9611 68.1495 65.2857	49.3668 46.6831 43.8418	85.0412 69.0821 66.2816	90.4754 88.6832 85.7338
Paper	AOS	79.23 70.58 67.89	53.12 49.87 46.32	88.12 73.65 70.21	91.78 89.12 86.87

Performance Evaluation Plots

The following plots compare the model's performance with results from the original PointPillars paper.

Detection Visualization

Below are example detection results on test images.

• Detection on Image 000080
• Detection on Image 000098
• Detection on Image 000138

Dataset

You can download the dataset from the following link:

🔗 Dataset Link

Execution Guide

To run this project, follow the steps below.

1. Download and Set Up the Dataset

The dataset needs to be downloaded separately and placed inside the dataset/ folder.

2. Install Dependencies

Install required dependencies using:

pip install -r requirements.txt

Then, set up the environment:

python setup.py develop

This installs the necessary packages and compiles any required modules.

3. Preprocess the Dataset

Before training, preprocess the dataset. If the dataset is inside the project folder:

python pre_process_kitti.py

If the dataset is stored elsewhere, specify its path:

python pre_process_kitti.py --data_root your_path_to_kitti

4. Train the Model

To train the model, use:

python train.py

For a dataset in a different location:

python train.py --data_root your_path_to_kitti

5. Run Inference on Test Images

To test an image, run:

python test.py --img_num 000002

This automatically finds the corresponding .bin, .txt, and .png files in dataset/demo_data/test/.