Skip to content
An attempt to create a script based on Konrad Koelzer's bundle import script that optimizes VSFM output for Blender
Branch: master
Clone or download
Pull request Compare This branch is even with soniquev8:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md
Sample Data.zip
import vsfm.py

README.md

VSFMBundle_to_Blender

An attempt to create a script based on Konrad Koelzer's bundle import script that optimizes VSFM output for Blender

The original script expected a very specific directory structure and file names, and also imported the bundle in what I feel is an in-efficient format.

The point cloud seems fine, but the original script wants to create a new camera and image plane for every image that was originally referenced in VSFM.

This approach takes up extreme amounts of memory and when using a large number of hi-res images (from VSFM), it can slow Blender to a crawl... not to mention the added drudgery of changing the active camera.

Goals

  1. Inspired by this tutorial: http://youtu.be/xjjW2yKjUDM my intent is to alter the script in order to import the referenced images as an image/movie sequence, 1 image per frame

  2. Recently discovered CloudCompare, an amazing bit of software that does precisely what I'm aiming for. However it has proven difficult to adapt their code into Python. As far as I can tell, my math is correct, but the points just aren't lining up the way they do in CC. They're working with the Bundle file whereas I'm working with the cameras_v2 file, but I believe I've compensated for all the differences. Sigh... this is a brain buster.

Advantages

I see the workflow of this script as basically importing the data from a VSFM export (bundle.rd.out) as well as the images from the 'data' subdirectory without having to re-arrange anything.
The user could then cycle through the images using left and right on the keyboard in order to 'connect the dots' of the point-cloud and model any objects as they see fit.

I've always found MeshLabs to be an obtuse intermediary with a very user-unfriendly interface, especially if a user (like me) is used to Blender. Also, many times I've attempted to use Meshlabs to create an actual mesh, the results have been less than staggering. This is especially true for rooms and unsequenced imagery.

Caveats

Unlike Meshlab, the current version of the bundle_importer script doesn't automatically texture anything or provide color data of any sort apart from the images themselves. Perhaps this functionality could be added in the future however it is not what I'm focusing on at the moment.

I pine for the day when VSFM/Meshlab/Boujou and photogrammetry functionality is integrated into Blender (hell, if my iPhone can do it with programs like Seene and 123DCatch, Why can't my computer???). But until then, my intent is simply to cut out the middle-man (Meshlab or the even more cumbersome Bundler or PPT) from the Blender photogrammetry workflow.

Any help would be much appreciated.

You can’t perform that action at this time.