[BUG] Invert after intensity normalisation

[qin-yu]

Describe the Bug (might not be a bug tho)

Hi @carsen-stringer!

I found three suspicious implementations in intensity normalization but could only identify two as potential bugs. I attempted to fix these two in #981, but I still wonder if the third issue should be addressed:

To invert the image so that dark cells become bright cells while the bright background becomes darker, the current implementation is:

cellpose/cellpose/transforms.py

Lines 666 to 667 in 84344b0

	if (tile_norm_blocksize > 0 or normalize) and invert:
	img_norm[..., c] = -1 * img_norm[..., c] + 1

essentially:

image = -1 * image + 1

However, image here is usually the output of a percentile-based normalization like normalize99() and is not strictly within the [0, 1] range. I guess most pixels are in [0, 1] so this inversion makes sense, but still some are outside [0, 1]. I haven't thoroughly thought this through. Is this a problem, or is it working as intended?

Many thanks!

[carsen-stringer]

thanks for looking into this, yes this is how it should work, like when not inverted, some pixels will lie outside the [0,1] range