Blender Photoscan Addon
Installation Guide
Compatibility
This is the official documentation for how to use
Blender Photoscan.
Note that this addon is only compatible with Windows (64-bit) and Blender 4.3!
There is a chance your computer is missing some external dependencies. If you happen to not have the
Microsoft C++ Redistributable then install both the x64 and x86 versions (we assume you are not using ARM64).
Most modern computers are powerful enough to run this addon. But keep in mind that it comes in two flavors:
Cuda Version: An accelerated build (using CUDA-enabled colmap). Use this only if you have an Nvidia GPU!
Non-Cuda Version: Slower build. Only use this version if you do not have an Nvidia GPU!
When you're done installing, quit and restart Blender!
If you run into permission issues, make sure to run Blender in administrator mode. This will allow Photoscan to read/write necessary files. It is also strongly recommended to save both the working .blend file and your dataset on your actual drive (not OneDrive, Google Drive, etc).
Parameters
Image Path - pointer to the folder that has all your images.
Quality - the quality of the dense reconstruction step. Directly affects speed!
Sparse - determines whether the sparse cloud is imported after calculation (it is always calculated under the hood).
Dense - determines whether the dense reconstruction is calculated at all (disable to skip this step).
Advanced Settings - exposes some extra parameters. For advanced users.
Variable Zoom - enable if your images have different focal lengths (otherwise Photoscan assumes all images are the same zoom-level).
Show Directory - opens the working directory as the calculation proceeds
Help - opens this webpage!
Console - shows console log during calculation. Strongly recommended!
Photoscan - proceeds with photoscanning (using images from the path and chosen settings).
Creating A Dataset
Messy or failed reconstructions are almost always the result of a bad dataset. Garbage in, garbage out. Always capture images the follow these guidelines:
High-Quality Images - your images should ideally be high resolution, sharp, and motion blur free.
Lots of Overlap - images should have lots of overlap (good rule of thumb is ≈ 80% between images). Photograph methodically - orbiting/moving only a little between images. Too few images (with low overlap) may make feature matching impossible.
Consistent Settings - Photoscan looks for similarities across the dataset. This means if lighting or white balance changes all of the sudden, identical features may no longer generate matches. If you zoom between images, make sure to enable variable zoom.
Valid Subject - similar to how changes in lighting throw off photogrammetry, the same applies to reflections and transmission. Avoid capturing subjects that are shiny/metallic or see-through.
Deep Focus - it's always good practice to capture extremely sharp images. Shallow depth-of-field is the enemy here. Set your aperture to f/11 or even higher if you can.
Consistent Scene - make sure your captured scene remains consistent. A car suddenly passing by, an object moved, or a shadow appearing can all throw off the photoscan.
Detailed Subject - even if your subject is non-glossy and opaque you may run into issues if it is largely featureless. The first part of the calculation involves feature detection. A blank white wall for example will have no features whatsoever. If your subject is largely featureless, a common trick is to cover it with baby powder.
Here is an example of a basic dataset that follows these rules. Note the choice of a non-glossy detailed object, consistent lighting, deep focus, etc:
Libraries
Photoscan is all about integrating photogrammetry right into Blender. As such it comes with the following packed libraries:
Colmap - for extracting and matching features (CUDA compatible - use an Nvidia card for even faster performance).
Glomap - for camera posing and sparse cloud generation (optimized library).
OpenMVS - for dense cloud calculation and mesh reconstruction.
