io.masks_flows_to_seg errors silently without saving seg npy file [BUG]

[davidhbrann]

Thanks for the nice software. I trained a model in the GUI and I was trying to now use a jupyter notebook to apply it to multiple images. I tried following the tutorial notebooks, and the model was running fine, but when I ran io.masks_flows_to_seg, I did not see any seg npy files. I'm using python 3.8.19 on windows and cellpose version 3.0.7.

https://github.com/MouseLand/cellpose/blob/509ffca33737058b0b4e2e96d506514e10620eb3/cellpose/io.py#L499

I believe the above line is the issue. imgs_restore=None but then [] * len(masks) always equals an empty list [] of len 0 (even though something like [None] * len(masks) would return a list of lists of len(masks)). This causing the enumerate command on line 501 to fail silently and no .npy https://github.com/MouseLand/cellpose/blob/509ffca33737058b0b4e2e96d506514e10620eb3/cellpose/io.py#L501 files are created.

I believe the same error is also in the test test_cyto2_to_seg which at the moment might also return without error but does not save the seg npy files.

I think changing the offending line to imgs_restore = [[] for _ in masks] might work.

The same error doesn't affect the GUI, which uses the _save_sets function instead.

As a reference one can see the differences in the following lines:

n = 3
masks = [1] * n
r1 = [] * n
r2 = [None] * n

# nothing is printed since r1 is len 0
for i, v in enumerate(zip(masks, r1)):
    print(i, r1)
# prints 1, 2, 3
for i, v in enumerate(zip(masks, r2)):
    print(i, r2)

(sorry, just noticed this is a duplicate of https://github.com/MouseLand/cellpose/issues/894)

[JulianoGianlupi]

I may have encoutered the same type of behavior for io.save_rois, empty zips. Either that or segmentation that was working on 2.2.3 stopped working on v3

[maorkn]

Same problem here. I've updated, and now, when I run:

    masks, flows, styles, diams = model.eval(imgs, diameter=None, channels=channels)
    io.masks_flows_to_seg(imgs, masks, flows, new_files, channels=channels, diams=diams)
    io.save_to_png(imgs, masks, flows, new_files)

I get the PNG files, but not *.seg.npy files without an error

[mrariden]

Should be fixed with #932