Added a pytorch implementation of OmniCV's to remove equilib package usage #2350
Conversation
| img: torch.Tensor, shape (batch_size, channels, height, width) | ||
| fov: int, field of view | ||
| theta: int, left/right angle | ||
| phi: int, up/down angle | ||
| hd: int, height of perspective image | ||
| wd: int, width of perspective image | ||
|
|
||
| return: torch.Tensor, shape (batch_size, channels, hd, wd) |
There was a problem hiding this comment.
Follow nerfstudio format for docstrings
| axis: torch.Tensor, shape (3,) | ||
| theta: torch.float | ||
|
|
||
| return: torch.Tensor, shape (3, 3) |
There was a problem hiding this comment.
Follow nerfstudio format for docstrings
There was a problem hiding this comment.
oh oops I'll fix that soon
|
|
||
| from nerfstudio.utils.rich_utils import CONSOLE, ItersPerSecColumn | ||
|
|
||
|
|
||
| # https://gist.github.com/fgolemo/94b5caf0e209a6e71ab0ce2d75ad3ed8 | ||
| def rotation_matrix(axis: torch.Tensor, theta: torch.float) -> torch.Tensor: |
There was a problem hiding this comment.
I think this functionality already exists -
nerfstudio/nerfstudio/cameras/camera_utils.py
Line 451 in 53c7c8e
There was a problem hiding this comment.
oh awesome, I'll use that
There was a problem hiding this comment.
so it turns out the functions were different enough so I just kept the new one but renamed it to decrease confusion. The new one implements Euler-Rodriquez rotation.
|
Looks good, but need to remove the library from the dependency list. |
done |
The equilib package was under a GPL so we couldn't use it in nerfstudio since nerfstudio is MIT. I just implemented the equirectangular to perspective method from this MIT licensed library in pytorch and then used it instead.