Description

Introduction
This training provides a practical and comprehensive journey into modern 3D vision and object detection using deep learning, focusing on PointNet-based architectures, voxelization approaches like VoxelNet, and the latest transformer-driven 3D perception models. Participants learn how to process point clouds, build detection pipelines, apply sensor fusion, and deploy real-time 3D detection systems across autonomous vehicles, robotics, AR/VR, and smart surveillance applications.

Prerequisites
Basic Python programming
Understanding of machine learning and deep learning concepts
Familiarity with NumPy and PyTorch or TensorFlow
Optional: Exposure to 2D computer vision concepts

Table of Contents
1. Introduction to 3D Vision & Deep Learning
 1.1 Evolution from 2D to 3D vision
 1.2 Point clouds, meshes and voxel representations
 1.3 Applications in autonomous vehicles, robotics, AR/VR

2. Point Cloud Processing Fundamentals
 2.1 LiDAR sensors and data formats
 2.2 Point cloud visualization tools
 2.3 Preprocessing: filtering, segmentation and downsampling

3. PointNet & PointNet++ Architectures
 3.1 Core design principles of PointNet
 3.2 Hierarchical feature learning in PointNet++
 3.3 Classification, segmentation and detection workflows

4. Voxel-Based 3D Detection with VoxelNet
 4.1 Voxelization and feature encoding
 4.2 Sparse convolutions for efficient 3D learning
 4.3 Implementing 3D detection with voxel networks

5. 3D Transformers for Object Detection
 5.1 Attention mechanisms for 3D data
 5.2 Transformer-based 3D detectors
 5.3 Comparing CNN, PointNet and Transformer architectures

6. Sensor Fusion Techniques
 6.1 LiDAR + camera integration
 6.2 Depth map fusion and BEV (Bird’s-Eye View)
 6.3 Multi-modal feature representations

7. Frameworks & Tooling for 3D Detection
 7.1 Open3D and PyTorch3D workflows
 7.2 MMDetection3D and Detectron3D
 7.3 ROS integration for robotics

8. Training, Optimization & Evaluation
 8.1 Dataset preparation and augmentation
 8.2 Metrics: mAP, IoU, KITTI/nuScenes standards
 8.3 Model compression and real-time deployment

9. Hands-On Projects
 9.1 Building a PointNet-based classifier
 9.2 Implementing VoxelNet for 3D object detection
 9.3 3D transformer detection pipeline with fusion

This training equips learners with the skills to design, train and deploy powerful 3D object detection models using PointNet, VoxelNet and transformers. By the end, participants gain hands-on experience and the confidence to build real-world 3D perception solutions across advanced AI-driven environments.