Error for min_size

[shan-stat]

Hi, thanks for developing the excellent program.

I tried to use the min_size argument in models.Cellpose.eval() function, but I think it did not work.
Please show the following results. Even though a mask consists of 82 pixels and the min_size argument is set to 100, the mask is not removed. Could you check this issue? Thanks.

---

masks, flows, styles, diams = model.eval(img, diameter=20, channels=channels, flow_threshold=0.7, cellprob_threshold=-2, resample=True, min_size=15)
processing 1 image(s)
time spent: running network 2.58s; flow+mask computation 2.25
estimated masks for 1 image(s) in 4.89 sec

TOTAL TIME 4.89 sec
masks.max()
301
len(np.where(masks==4)[0])
82
len(np.where(masks==4)[1])
82

masks, flows, styles, diams = model.eval(img, diameter=20, channels=channels, flow_threshold=0.7, cellprob_threshold=-2, resample=True, min_size=100)
processing 1 image(s)
time spent: running network 2.58s; flow+mask computation 2.25
estimated masks for 1 image(s) in 4.88 sec

TOTAL TIME 4.88 sec
masks.max()
301
len(np.where(masks==4)[0])
82
len(np.where(masks==4)[1])
82

---

[GFleishman]

Just curious why this issue was closed? I have some questions about min_size as well; though have not observed this specific problem.

[ashishUthama]

I notice the same issue - Explicitly set min_size is ignored for 2D (defaults to 15).

Traced this to the following call where min_size is not propagated to dynamics.compute_masks in the 2D case:

cellpose/cellpose/models.py

Line 652 in e559dae

outputs = dynamics.compute_masks(dP[:,i], cellprob[i], niter=niter, cellprob_threshold=cellprob_threshold,