The new cyto2 network cannot be used from command line

[thosoo]

in __main__.py:

if not (args.pretrained_model=='cyto' or args.pretrained_model=='nuclei'):
                cpmodel_path = args.pretrained_model
                if not os.path.exists(cpmodel_path):
                    logger.warning('model path does not exist, using cyto model')
                    args.pretrained_model = 'cyto'

When adding --pretrained_model cyto2 in the command line, it will obviously not work.

However, I have not tried with the full path to the models yet.

[giocard]

The issue is still not closed. I could use the cyto2 from command line by modifying

cellpose/cellpose/__main__.py

Line 142 in 4b1eaa3

if args.pretrained_model=='cyto' or args.pretrained_model=='nuclei':

in the same way it has been already done for line 130.

cellpose/cellpose/__main__.py

Line 130 in 4b1eaa3

if not (args.pretrained_model=='cyto' or args.pretrained_model=='nuclei' or args.pretrained_model=='cyto2'):

[carsen-stringer]

I'm confused, does the github version work for you? if so I'll close the issue

[giocard]

No, it doesn't. When running from command line, if the diameter is set to zero the program stops at line 147

cellpose/cellpose/__main__.py

Line 147 in 4b1eaa3

diameter = model.diam_mean

because it thinks that is a model provided by the user, and therefore it expects the diameter.
What I was trying to say in my previous post is that I had to fix line 142 by adding or args.pretrained_model=='cyto2' to make it work also under this condition. If I set the diameter to any positive value, the program runs fine.

[giocard]

I forgot to mention that in any case, even with a model provided by the user, the program would fail if the input diameter is set to zero, because of the following error at line 147

UnboundLocalError: local variable 'model' referenced before assignment

[carsen-stringer]

ah good catch, please try now

[giocard]

I did a few tests and it seems to be fixed now. Thanks!

[carsen-stringer]

thanks for your help!