depth_aware

dnois.optics.depth_aware(obj: Tensor, mask: Tensor, psf: Tensor, pad: int | tuple[int, int] | str = 'linear', occlusion_aware: bool = False) → Tensor

Depth-aware image formation model. The obj (sharp image) is first segmented into depth planes w.r.t. depth according to mask. Then the portion in each plane is convolved with the PSF of corresponding depth. Final image (blurred image) is the superposition of all the planes.

Note that both obj and psf can be real representing incoherent imaging and complex representing coherent imaging. If just one of them is complex, the other will be cast as complex. The blurred image is real if and only if they are both real.

Smaller indices in \(D\) dimension in mask and psf are expected to represent smaller depths.

Parameters:

• obj (Tensor) – A tensor of shape \((\cdots,H_o,W_o)\).

• mask (Tensor) – A tensor of shape \((\cdots,D,H_o,W_o)\).

• psf (Tensor) – A tensor of shape \((\cdots,D,H_p,W_p)\). PSF should be normalized, see dnois.optics.norm_psf() for details.

• pad (int, tuple[int, int] or str) – Padding width used to mitigate aliasing. See dnois.fourier.dconv2() for more details. Default: 'linear'.

• occlusion_aware (bool) – Whether to use the occlusion-aware image formation algorithm [1]. Default: False.

Returns:

A tensor of shape \((\cdots,H_o,W_o)\).

Return type:

Tensor

[1]

Ikoma, H., Nguyen, C. M., Metzler, C. A., Peng, Y., & Wetzstein, G. (2021, May). Depth from defocus with learned optics for imaging and occlusion-aware depth estimation. In 2021 IEEE International Conference on Computational Photography (ICCP) (pp. 1-12). IEEE.