Hi, thanks for releasing NeoVerse and the code. I have a question about the global camera trajectory mode.
I would like to use an official camera trajectory provided by the dataset as input to the global mode. My understanding is that global directly treats the provided OpenCV c2w matrices as target camera poses. However, NeoVerse’s reconstruction uses its own coordinate system and scale, so an external trajectory cannot be used directly even if the axis convention is consistent.
Would the correct workflow be:
Obtain NeoVerse’s predicted camera poses for the input frames.
Estimate a Sim(3) alignment between the predicted poses and the corresponding official camera poses.
Transform the official trajectory into NeoVerse’s coordinate system.
Pass the aligned trajectory to global mode.
Is this the intended usage? If so, is there a recommended way to obtain NeoVerse’s predicted camera poses for the alignment?
Thanks for your help!
Hi, thanks for releasing NeoVerse and the code. I have a question about the global camera trajectory mode.
I would like to use an official camera trajectory provided by the dataset as input to the global mode. My understanding is that global directly treats the provided OpenCV c2w matrices as target camera poses. However, NeoVerse’s reconstruction uses its own coordinate system and scale, so an external trajectory cannot be used directly even if the axis convention is consistent.
Would the correct workflow be:
Obtain NeoVerse’s predicted camera poses for the input frames.
Estimate a Sim(3) alignment between the predicted poses and the corresponding official camera poses.
Transform the official trajectory into NeoVerse’s coordinate system.
Pass the aligned trajectory to global mode.
Is this the intended usage? If so, is there a recommended way to obtain NeoVerse’s predicted camera poses for the alignment?
Thanks for your help!