substitute my own depth and camera predictions as initialization

[RuibinMa]

Dear authors,

I'm wondering if I can initialize the depth values and camera poses differently? Currently I have some predicted depth map and camera poses from other methods and want to make use of the direct local windowed optimization in DSO. Could you make some suggestions like which lines of code I should start with?

Currently, for camera poses, I have changed the code in FullSystem::trackNewCoarse. The modification basically replace "fh_2_slast" with my predicted relative camera transformation.

For depth maps, my current idea is to modify lines in CoarseInitializer wherever it tries to assign a uniform initial depth value to "idepth" and "iR". (BTW, what's the difference between them idepth and iR?)

Do you think my modifications are on the right track? Any suggestions will be greatly appreciated!!

Thanks!

[vonzy]

I suggest you don't try to modify CoarseInitializer, the readability is not so good. You can write your own initializer.