Here is the code that was used for the paper
(with some minor changes).
And here is a small data set to test it with.
Most of it is in matlab but in addition the following is needed:
The five point solver, Lowe's SIFT implementation , VLfeat and Mosek .
To run it, run the script reconstruct_scene.m. You will have to change the paths in corner/reconstr_setup.m to make it work.
The code is supplied with absolutely no guarantees whatsoever and probably full of bugs. If you use it cite the paper.
In the near future I am planning to update the rotation averaging routine which is not very good for large data sets it seems.
Datasets:
These are the datasets from the paper. Due to size limitations I cannot put the images here.
Instead I have put the tracked image points (obtained after step 2 in the algorithm) and the final reconstruction.
If you would like to have the images you will have to contact me.
The image point file contains a structure with the fields pointnr, points and index.
The points-field contains the two-dim. coordinate for the point and the cooresponding position in the index field
contains the number of the corresponding 3D point. The field pointnr is the total number of 3D-points.
Carl Olsson, Olof Enqvist,
The Arch of Triumph, Paris.
