2D to 3D Image Generation using Depth Estimation

This project converts a single 2D image into a 3D point cloud using monocular depth estimation powered by the GLPN deep learning model. The 3D scene is then visualized using Open3D.

---

📌 Project Overview

This Python-based system:

• Loads a single 2D image
• Applies the GLPN (Global-Local Path Network) model from Hugging Face to predict depth
• Converts depth + RGB into an RGBD image
• Builds a 3D point cloud using Open3D
• Visualizes or exports the point cloud

---

✅ Real-World Applications

🎮 AR/VR & Game Dev

• Quickly convert 2D concept art or objects to 3D form

🤖 Robotics

• Enable depth perception using just RGB cameras

🎨 Creative Media

• Enhance flat images with pseudo-3D visuals or effects

---

🔧 Why This Approach?

• Pretrained Model: Leverages vinvino02/glpn-nyu for accurate depth prediction
• Offline: All computations run locally
• Flexible: Works on any standard image format (JPG, PNG)
• Lightweight: Requires no GPU (optional but beneficial)

---

🛠️ Requirements

• Python 3.8+
• PyTorch
• Open3D
• Transformers
• Pillow
• Matplotlib

Installation:

pip install torch torchvision transformers pillow matplotlib open3d

📽️ Demo

---