MultiViewRenderer

License

Note

Based on the Open3D codebase context,the rendering type is categorized as ray-casting, not ray-tracing or rasterization. The terminology can be confusing since Open3D uses Embree (a ray-tracing library) but implements ray-casting functionality. The key distinction lies in its application: ray-casting is used for geometric queries and volume rendering, not for photorealistic ray-tracing.

Table of Contents

• MultiViewRenderer
  • Table of Contents
  • Directory explanation
  • Install
  • Usage
    • Example
    • Output Files Documentation
      • File Structure
      • File Details
        • Color Images (color_*.png)
        • Depth Maps (depth_*.tiff)
        • Camera Parameters (camera_params.npz)
        • Configuration Log (config.log)

Install

git clone https://jahad9819jjj/multi_view_rendr
cd multi_view_rendr
pip install -r requirements.txt

Usage

Example

Use The Stanford 3D Scanning Repository's bunny file as an example.

cd multi_view_rendr/data
wget http://graphics.stanford.edu/pub/3Dscanrep/bunny.tar.gz
tar -zxvf bunny.tar.gz
python3 scripts/main.py --debug

Output Files Documentation

This document explains the output files generated by the multi-view rendering pipeline.

File Structure

dump/
├── color_000.png # RGB color image (View 0)
├── color_001.png # RGB color image (View 1)
├── ...
├── depth_000.tiff # Depth map (View 0)
├── depth_001.tiff # Depth map (View 1)
├── ...
├── camera_params.npz # Camera parameters archive
└── config.log # Configuration log file

File Details

Color Images (color_*.png)

• Format: PNG (Portable Network Graphics)
• Dimensions: [height]x[width]x3 (RGB)
• Content:
  • Rendered RGB views of the point cloud from different camera angles
  • Images are saved in OpenCV's BGR color format
• Naming Convention:
  • color_XXX.png where XXX is 3-digit zero-padded view index
  • Example: color_002.png = 3rd captured view

Depth Maps (depth_*.tiff)

• Format: TIFF (Tagged Image File Format)
• Dimensions: [height]x[width] (single channel)
• Content:
  • 32-bit floating point depth values
  • Depth in meters from the camera plane
  • Invalid pixels marked with 0.0
• Usage Tips:

  import cv2
  depth = cv2.imread("depth_000.tiff", cv2.IMREAD_UNCHANGED)

Camera Parameters (camera_params.npz)

Format: NumPy compressed archive Contents:

{
    'intrinsics': [3x3 intrinsic matrices],  # List of camera intrinsics
    'extrinsics': [4x4 extrinsic matrices]   # List of camera poses
}

import numpy as np
data = np.load('camera_params.npz')
intrinsics = data['intrinsics']  # Shape: (N,3,3)
extrinsics = data['extrinsics']  # Shape: (N,4,4)

Configuration Log (config.log)

Format: Text file Contents:

[Timestamp] YYYY-MM-DD HH:MM:SS
Input File: /path/to/input.ply
Output Directory: /path/to/output
Camera Configuration:
  - Resolution: 1920x1080
  - Focal Length: (1372.45, 1350.36)
  - Principal Point: (1049.30, 593.64)
View Parameters:
  - Number of Views: 12
  - View Distribution: spherical
  - Radius Range: [1.5, 2.5] meters