Dataset preparation (camera pose)

vislearn / dsacstar

DSAC* for Visual Camera Re-Localization (RGB or RGB-D)

BSD 3-Clause "New" or "Revised" License

235 stars 36 forks source link

Dataset preparation (camera pose) #9

Closed YaroslavShchekaturov closed 2 years ago

YaroslavShchekaturov commented 3 years ago

HI!

Am I right that if I want to use camera pose extracted from the blender ( obj.matrix_world) I need to multiply it by diag(1 -1 -1 1) to follow the OpenCV convention?

Thank you!

ebrach commented 3 years ago

I'm not sure what convention Blender uses, but you are right about DSAC* using the OpenCV convention. If Blender uses the OpenGL convention, you might have to flip twice: Flip second and third column, then flip second and third row. (Result of multiplying diag(1 -1 -1 1) from left and from right.)

Let me know whether you solved the issue.

Best, Eric

YaroslavShchekaturov commented 3 years ago

Thank you!

I solved this issue when I simply saw this picture.

So I did it like this

Best, Yaroslav