Structure from motion projects

December 20th, 2010 by Henri Leave a reply »

I’ve introduced my tracking algorithm in the previous post. One of the issue I have is that the point cloud generated by my SFMToolkit (using Bundler) is not always accurate. This is a list of structure from motion projects alternative I’m interested in:

Building Rome in a Day:

Project home is using Bundler (GPL)

Building Rome on a Cloudless Day:

Project home | Source code (Non-profit license, I’ve ported their source to windows)

Samantha:

Project home (I’ve contacted them without response but they said that they were going to release the source code: check at 28:50)

Samantha Bundler

PhotoSynth:

Website – Microsoft closed-source SFM application: check out my PhotoSynthToolkit

PhotoSynth Bundler

ETH-V3D Structure-and-Motion software:

Project home with source code (GPL, I’ve partially ported it to windows)

Simple Sparse Bundle Adjustment:

Project home with source code (LGPL, I’ve ported it to windows)

A multi-stage linear approach to structure from motion:

Project home | paper

Results from the paper of LinearSFM (Microsoft Research)

This list is not exhaustive, I’ve seen other projects (Efficient Large Scale Multi-View Stereo for Ultra High Resolution Image Sets: not sure how it is related to ETH-V3D Structure-and-Motion software)

Share
Advertisement

11 comments

  1. Cesar Lopez says:

    So which one seems the most viable for your SFMtoolkit?

  2. admin says:

    @Cesar Lopez: you’re right I should have commented all project I’m interested in:

    • -Building Rome in a Day: just mentioned for reference (using Bundler)
    • -Building Rome on a Cloudless Day: the source code does not contain the sfm solution, but it contains Cuda Ransac and Cuda Gist descriptor with k-medoids and k-nearest neighbors
    • -Samantha: they still don’t have released their source code but it sounds very promissing
    • -PhotoSynth: I’m trying to triangulate my own feature tracks using PhotoSynth camera position
    • -ETH-V3D Structure-and-Motion software: I’m currently trying to port this to windows, seems to be the most complete Bundler alternative
    • -Simple Sparse Bundle Adjustment: this is not really a sfm solution, it only does the bundle adjustment but I may use this with OpenSynther
    • -A multi-stage linear approach to structure from motion: this is a very great paper achieving impressive speed-up compared to Bundler and is “drift-free”

    So there is not only one alternative to Bundler this is why I’m interested in all this great projects.

  3. Josh Harle says:

    Hi Henri,

    Wow; the PhotoSynth/Bundler comparison is quite an indictment of Bundler results! :)

  4. Pierre says:

    ETH-V3D Structure-and-Motion software: I’m currently trying to port this to windows, seems to be the most complete Bundler alternative => Will not be easy. Need LP_Solve, Cuda…

  5. admin says:

    @smoggy: thanks a lot for this link!

    @Josh Harle: I know that the PhotoSynth/Bundler comparison it’s now really fair… But this is the result I got from my church dataset. And as I’m using the sparse point cloud for tracking you understand why I’m looking for Bundler alternative.

    @Pierre: actually I already have it working ;) :

    SfM package:
    - all binaries are working

    CycleInference package:
    - nothing (not tried yet)

    The windows port wasn’t easy due to the number of dependencies (+ some cuda rewrite to be compatible with visual studio): I’ll share it once I’ll have the scripting complete too.

  6. Pierre says:

    Nice job to make it works !
    It seems that the dependency over LP_solve is not used in the code in fact… Just the includes are present …
    I’m quite impatient to test it.
    If you want some help for the cycle inference. Tell me ;)

  7. admin says:

    @Pierre: Thanks! I’m officially on vacation today (no coding allowed during holiday…)! So you’ll have to wait 2011 to get the visual studio solution. Hopefully it won’t be long! BTW if you try to compile the solution with V3DLIB_ENABLE_LPSOLVE you’ll see that LP_solve is needed. Concerning the CycleInference package: I think that libDAI is the only dependency missing (I should be able to compile it myself, but thanks for the help!).

  8. Please take at a look at http://www.visualsize.com. Visualsize conducted unbiased comparison studies against Bundler, Bundler+PMVS2, and Autodesk Photofly, and demonstrated that their 3D software outperforms all others based on a similar SfM principle by a significant margin. Visualsize provides more than 100 3D models of all kinds of objects using nothing but uncalibrated digital photos.

  9. admin says:

    @marriedwchildren: I’m aware of your product but I didn’t see any publication nor downloadable product than I can test with my own dataset: this is why I don’t have included it on the list of SFM project I’m interested in. But you seems to have indeed a very accurate SFM solution. I just have one concern about your unbiased comparison: why do you have used so small picture in the PhotoSynth comparison?

  10. bryan says:

    i’m interesting with Building Rome on a Cloudless Day but i’m not found source code. Does the source code actually be shared? (sorry for my bad english)

Leave a Reply