Processing WSI - spatial constrains order

[BBolosSierra]

Hi there,

Thanks for your work, I am using both FPS and PathDino, and it is really great so far.
I actually have a question about the order of the FPS steps.

In the paper, the spatial constrains are applied to $S$ (sampled after KDE):

$$ s_i, s_j \in S, \sqrt{(s_{i,x} - s_{j, x})^2 + (s_{i, y} - s_{j, y})^2} \geq e_{min} $$

However, in the code, it looks like the spatial constrains are applied to $P$, the potential patch locations. Subsequently, the KDE is used for sampling:

https://github.com/KimiaLabMayo/PathDino/blob/main/FPS%20code/FPS_helper.py#L128

`

    patch_locations, image_with_contours, image_with_rectangles = get_patch_locations(tissue_mask, cthumbnail, mask_hratio, mask_wratio, tissue_percent)
    
    min_distance = mask_hratio * 2  # Minimum distance between points

    filtered_patch_locations = []
    for (x, y), _ in patch_locations:
        if is_far_enough((x, y), filtered_patch_locations, min_distance):
            filtered_patch_locations.append((x, y))

    filtered_patch_locations = get_sampled_points_density_proportional_KDE(filtered_patch_locations, returnSamples)`

I was wondering if there is a reason for reversing the paper formulation order to do spatial constrains to $P$ first, and then KDE to get $S$. May this be a bug or a mistake in the paper?

Thank you so much in advance,
Best,
Bego