I have question about what position and rotation represent in cameras.json, i.e. C2W or W2C?
I understand that the original code to output cameras.json is in Gaussian Splatting repo and it writes W2C. However, I have the same doubt as this issue that cameras.json actually represents C2W.
This would affect the rendered view in PhysGaussian repo (and probably follow-up works). By default, the config file only samples one default_camera_index to render a view in get_camera_view, but when I try to sample multiple camera poses, it doesn't transform as expected.
I plotted the camera poses when assuming cameras.json represents C2W and W2C as follows. C2W seems to make more sense to me. Could you please clarify on that?
Hi, thanks for the great work!
I have question about what
positionandrotationrepresent incameras.json, i.e. C2W or W2C?I understand that the original code to output
cameras.jsonis in Gaussian Splatting repo and it writes W2C. However, I have the same doubt as this issue thatcameras.jsonactually represents C2W.This would affect the rendered view in PhysGaussian repo (and probably follow-up works). By default, the config file only samples one
default_camera_indexto render a view inget_camera_view, but when I try to sample multiple camera poses, it doesn't transform as expected.I plotted the camera poses when assuming
cameras.jsonrepresents C2W and W2C as follows. C2W seems to make more sense to me. Could you please clarify on that?