camera pose refinement starting from relatively correct initial poses instead of identity transformation matrices

[tofis]

Thank you @chenhsuanlin for sharing this impressive work. I wanted to ask you whether it is possible to initiate BARF training from initial camera poses instead of identity matrices changing the configuration only or I have to adapt the code.

[chenhsuanlin]

Hi @tofis, it is definitely possible (as was in the Blender data experiments), but there would be a bit of code modification involved. The pose variable up to this line would be the initial poses (ground truth + noise for Blender and identity for LLFF), followed by the correction pose_refine. So you just have to make sure your initial camera poses are assigned to pose.
Hope this helps!

[tofis]

Thanks for the prompt reply! Indeed, that does the job!