Documentation
If you've ever used programs like RealityCapture, Metashape, Polycam, etc then you're familiar with Photogrammetry. Photos in - 3D model out. The basic pitch is that we should be able to do this directly inside Blender!
This is a flexible "plug n' play" solution.
These are all basic outputs using the Medium preset:
Here is a sparse proxy model with only a 10 image dataset:
Here is a demo of how to use the addon:
I wanted both the sparse reconstruction step along with the dense meshing step to be as fast as possible (note that photogrammetry in general is notoriously slow to begin with). For those with Nvidia GPUs, you will get Cuda accelerated reconstruction. I also used the fastest libraries I could find for feature extraction, matching, reconstruction, densifying, and meshing. For those interested I used (Colmap, GloMap, and OpenMVS).
There are three presets - Low, Medium, and High. Each outputting progressively more detail (at the cost of computation time).
This addon does not calculate textures!
Smoothing (post-import) not yet implemented.
Decimation (post-import) not yet implemented.
Blender Photoscan Addon
If you've ever used programs like RealityCapture, Metashape, Polycam, etc then you're familiar with Photogrammetry. Photos in - 3D model out. The basic pitch is that we should be able to do this directly inside Blender!
This is a flexible "plug n' play" solution.
Results
These are all basic outputs using the Medium preset:
Here is a sparse proxy model with only a 10 image dataset:
Demo
Here is a demo of how to use the addon:
Optimized For Speed
I wanted both the sparse reconstruction step along with the dense meshing step to be as fast as possible (note that photogrammetry in general is notoriously slow to begin with). For those with Nvidia GPUs, you will get Cuda accelerated reconstruction. I also used the fastest libraries I could find for feature extraction, matching, reconstruction, densifying, and meshing. For those interested I used (Colmap, GloMap, and OpenMVS).
There are three presets - Low, Medium, and High. Each outputting progressively more detail (at the cost of computation time).