-
signal_strength
- Between 1 and 255 scaling of image intensity before poisson noise
-
psf_scale
- gaussian psf width in sigma (I think)
-
coin_flip_bias
- bionomial odds
-
thinning_type:
- Split image into training (T) and validation (V)
- poisson:
- Bionomial splitting
- spatial:
- All spatial thinning methods split the image into two
- spatially rather than per pixel.
- To make these images work with the rest of the code
- they need to be stretched in the thinned axis
- spatial_interlaced
- zero out signal every other pixel
- spatial_interpolated
- smooth linear interp
- spatial_repeated:
- take left pixel and copy to right pixel
-
metrics:
- gt_error_l[1/2]:
- l[1/2]_norm between x and ground_truth
- gt_error_ssim
- structural similiarity
- Rnrm_[VAR]:
- Sum of normalised residuals between T and V
- log_liklihood_[VAR]
- Liklihood between Ax and [VAR]
- gt_error_l[1/2]: