[Soc-2016-dev] Weekly Report #6, Multiview Reconstruction

[Tianwei Shen]

Hi all,

What I did this week:
- The primary work of this week is bug fixing. Following my work last week, the multi-view reconstruction is now able to run and return the re-projection error.

- Add varying intrinsics APIs. mv::Reconstruction has a vector that saves camera intrinsics. I added a map data field from camera pose index to camera intrinsics index. For the current implementation, I will assume that the frames in the same clip will share a same set of camera intrinsics. But in the future, we may support varying focal lengths so that the frames in the same clip will have different camera index mapping. So this varying intrinsics APIs will benefit us in the long run.

- Remove some std::cout logging information in libmv and revert to glog for error message logging, format libmv code to google code style (mainly replace tabs with 2 spaces).

- My mentor (Sergey) helped me solve correspondence saving/loading issues, many thanks. I also got to know a bit of blender loader stuff. 

- Improve track correspondence adding. If the same correspondence has been added, give an error message. If a conflict correspondence is added, which means the same track in one view is linked to more than one track in other views, give an error message. 

What I will do next week:
- Pass libmv result back to blender and visualize camera motion in the 3D view of blender. 

- Optimize the camera solving pipeline and minimize the re-projection error of multi-view reconstruction. 


Thanks,
Tianwei