Multi-GPU parallel computing

[Kuni05]

Hi!

Using demo_colmap + ba, but the memory usage is too high. I would like to know if there is a way to conduct multi-GPU parallel computing?

Thanks.

[jytime]

Hey I think it is definitely possible but did not try it before

[paradise-wine]

@Kuni05
Hi, I met the same problem, did you solve it?

[Kuni05]

@Kuni05
Hi, I met the same problem, did you solve it?

No.

[xiaoya27]

Hi! I’ve implemented support for multi-GPU parallel computing to address this issue.
You can see the proposed fix in this pull request #445

Please let me know if you have any feedback or if further adjustments are needed:)

[KKjiaming]

Hi! I’ve implemented support for multi-GPU parallel computing to address this issue. You can see the proposed fix in this pull request #445

Please let me know if you have any feedback or if further adjustments are needed:)

Thank you for sharing. However, I think VGGT relys on multi-perspective contexts. If the input data is only fragmented and only a part of the perspective is seen each time, the scale/pose consistency is more likely to deteriorate. It seems that it's quite difficult to write multi-card inference unless the model architecture is split?