Transform the world coordinate of the camera pose

[glennliu]

Hi @ShunChengWu

I want to transfer the global coordinate of the pose. In the Retrieve function of the "dataloader_3rscan.cpp", I add a steady transformation to the loaded camera pose. However, I found the transformed segments map is still aligned with the original segments map. And the color of the surfels is significantly twisted.

Original surfels:

Transformed surfels:

In the example above, I transformed the pose to +0.5m in the height direction. And you can see many of the textures are fused into surfels that higher than they are supposed to be. It looks like the depth image and the rgb image are not integrated with the same camera pose. Is there any part about the pose transform is hardcoded?

I was doubting is it because the rendered depth map is not transformed. But it looks like the depth is rendered on the retrieved pose, which is also loaded from the "dataloader". So, the rendered depth should be consistent with the coordinate transformation.

Thanks! Chuhao

[glennliu]

It turns out it is because of the render function. I shouldn't transform the pose used to render depth. Otherwise, the rendered depth is not consistent with my transformation. It is solved now. Regards