Generate Learning-based Multi-view Stereo dataset using Blender Python API

As it states in Blender's Python API Overview

Blender has an embedded Python interpreter which is loaded when Blender is started and stays active while Blender is running.

This repository provides ability to generate synthetic datasets using ONLY Blender executable without any additional dependency.

Script was tested on

❯ blender.exe --background --python-expr "import sys; print(sys.version)"
Blender 4.2.9 LTS (hash a10f621e649a built 2025-04-15 01:46:41)
3.11.7 (main, Jun 11 2024, 12:31:01) [GCC 11.2.1 20220127 (Red Hat 11.2.1-9)]

Blender quit

---

Examples

See help

Blender 4.2.9 LTS (hash a10f621e649a built 2025-04-15 01:46:41)
usage: blender [-h] model_path output_directory [number_of_renders]

Script for generating rendered images of a 3D model using Blender, driven by config.toml.

positional arguments:
  model_path         Path to the 3D model file (OBJ, FBX, glTF, GLB).
  output_directory   Directory where the rendered images will be saved.
  number_of_renders  Number of images to generate per run.

options:
  -h, --help         show this help message and exit
  --dump-config      Dump user configuration as python dict and exit.
  --skip-render      Setup everything and skip render.

1. Export monkey mesh Suzanne from blender as .glb (or download it here)

2. To generate one image into output directory use

   blender --background --python generate-batch.py -- suzanne.glb ./output

3. See result in output directory

   

4. To generate 16 images from random locations into output directory use

   blender --background --python generate-batch.py -- suzanne.glb ./output 16

5. See result in output directory¹

   

---

Configuration

Script provides extensive configuration via user defined config.toml, see config.default.toml for available parameters.

To dump parsed config and exit the program use

blender --background --python generate-batch.py -- --dump-config

---

Plugins

This project ships with a config‑driven plugin system that lets you bolt new functionality onto the rendering pipeline without touching the main script.

See available plugins and more in PLUGINS.md.

---

Compatibility

The compat/ directory contains helper scripts to restructure the output of this Blender dataset generator into formats compatible with popular neural rendering and 3D reconstruction frameworks:

Available Converters:

1. COLMAP - Converts to standard COLMAP structure (images/, sparse/)

   python compat/to_colmap/to_colmap.py <input_dir> <output_dir>

2. NSVF - Prepares data for Neural Sparse Voxel Fields pipeline

   python compat/to_nsvf_dataset/to_nsvf_dataset.py <input_dir> <output_dir>

3. IDR - Formats data for Implicit Differentiable Renderer requirements

   python compat/to_idr_dataset/to_idr_dataset.py <input_dir> <output_dir>

---

FAQ

See FAQ.md

Footnotes

1. this grid was made with this ↩