Upsampling of Point Clouds

Overview

Point Clouds are 3D data representations captured from LiDAR sensors, 3D scanners, or depth cameras. They are widely used in autonomous driving, robotics, VR/AR, and 3D object reconstruction.

This project explores point cloud upsampling using Deep Learning architectures like PointNet and PointNet++. The goal is to enhance the resolution of sparse point clouds, improving their quality for classification, recognition, and reconstruction tasks.

---

What is Point Cloud Upsampling?

Upsampling point clouds involves increasing point density while preserving object structure. This is essential for:

• 3D Object Classification & Recognition 📦
• LiDAR Data Processing for Autonomous Vehicles 🚗
• Medical Imaging & 3D Reconstruction 🏥
• Virtual Reality (VR) & Augmented Reality (AR) 🎮

PointNet & PointNet++ are deep learning models that process raw point cloud data directly, making them effective for upsampling and classification.

---

Tech Stack

• Programming Language: Python
• Libraries & Frameworks:
  • TensorFlow/PyTorch – Deep learning framework for training PointNet
  • NumPy – Efficient numerical computations
  • Open3D – Processing and visualization of point clouds
  • Matplotlib – 3D visualization

---

Implementation

1. Loading the Point Cloud Data

   • Reads a .ply/.pcd file using Open3D.

2. Processing with PointNet/PointNet++

   • Extracts geometric features.
   • Classifies objects into different categories.
   • Upsamples point clouds using deep learning.

3. Visualization

   • Displays before and after upsampling results.

---

📊 Results & Visualization

---