Related Experiment Video
Updated: Aug 23, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
606
3D Vehicle Detection and Segmentation Based on EfficientNetB3 and CenterNet Residual Blocks
1St. Petersburg Federal Research Center of the Russian Academy of Sciences, SPC RAS, 199178 St. Petersburg, Russia.
Sensors (Basel, Switzerland)
|October 27, 2022
Summary
This study introduces a novel two-stage method for 3D vehicle detection and segmentation using EfficientNetB3. The approach achieves state-of-the-art performance, outperforming existing solutions in 6DoF error.
Area of Science:
- Computer Vision
- Machine Learning
- Robotics
Background:
- Accurate 3D vehicle detection and segmentation are crucial for autonomous driving systems.
- Existing methods face challenges in precise localization and pose estimation in complex scenes.
Purpose of the Study:
- To develop a robust two-stage framework for 3D vehicle detection and segmentation.
- To improve the accuracy of 3D localization, pose estimation, and segmentation masks for vehicles.
Main Methods:
- A two-stage approach combining EfficientNetB3 with multiparallel residual blocks for initial 3D localization and pose estimation.
- A second stage using EfficientNetB3 for image recognition on cropped vehicle images.
- Utilizing predefined 3D models and transformation matrices for final 3D bounding box and segmentation mask generation.
Main Results:
- The proposed method achieved superior performance on the ApolloCar3D dataset.
- Outperformed all previously published solutions in terms of 6 degrees of freedom error (6 DoF err).
Conclusions:
- The two-stage EfficientNetB3-based framework offers a highly effective solution for 3D vehicle detection and segmentation.
- The method demonstrates significant advancements in accuracy for autonomous driving perception tasks.
