search

facebookresearch / vggsfm

VGGSfM: Visual Geometry Grounded Deep Structure From Motion
Other
907 stars 67 forks source link

Utilizing known camera intrinsics #64

Closed ojh6404 closed 1 month ago

ojh6404 commented 1 month ago

Thanks for great work.

I'm trying to use this work for camera pose estimation and I have a question. Is there anyway to utilize camera intrinsics (fx,fy,cx,cy) that I already know?

jytime commented 1 month ago

Hey, I believe the answer is yes. If you’re comfortable using them as initial estimates and then refining them during the bundle adjustment stage, you can simply replace the input intrinsics in the triangulator with your own data, as shown here:

https://github.com/facebookresearch/vggsfm/blob/6acf1f10f684600125a678ee466272c9eea6d58a/vggsfm/models/triangulator.py#L89

If you need the output intrinsics to match your input exactly, it’s possible, but I’ll need to explore how we can freeze the intrinsics during the bundle adjustment process.

ojh6404 commented 1 month ago

Thanks for your kind answer. I will give it a try.