[BUG] Mask values exceed 65535 when `metric_intersection_over_union()`

[Chenchao-is-CC]

Before you fill out this form:

Did you review the FAQ?

• Yes, I reviewed the FAQ on the [Cellpose ReadTheDocs](https://cellpose.readthedocs.io/en/latest/faq.html)

Did you look through previous (open AND closed) issues posted on GH?

• Yes, I [searched for related previous GH issues](https://github.com/MouseLand/cellpose/issues?q=is%3Aissue).

---

Describe the bug

When running 3D stitching on a large image stack (1682, 2693, 2482), the combined mask values in metrics_intersection_over_union() exceed uint16 (65535), causing integer overflow. Currently, the stitched mask array is initialized as uint16, which cannot accommodate large label counts.

---

Expected behavior

It would be helpful if Cellpose could automatically determine an appropriate dtype for stitched masks, promoting to int32 (or larger) when necessary, to prevent overflow. I would be very grateful if you could fix this bug.

---

Screenshots

---

Additional context

Automatic dtype promotion would prevent overflow issues in large 3D datasets and simplify downstream analysis.

[a3klein]

Hey, I have been running into the same issue. My suggested fix for this will be something along the lines of:

in cellpose/utils.py in the stitch3D function right before line 547 I would add the overflow check.

...
ino = np.nonzero(iou.max(axis=1) == 0.0)[0]

### Insert check for runtime overflow error
if istitch.dtype == "uint16" and (mmax + len(ino) + 1) > np.iinfo("uint16").max:
    istitch = istitch.astype("uint32")
    mmax = mmax.astype("uint32")
    if isinstance(masks, list): 
        masks = [mask.astype("uint32") for mask in masks]
    else: 
        masks = masks.astype("uint32")
    print("Some warning about increasing the dtype of the masks object")
### end check

istitch[ino] = np.arange(mmax + 1, mmax + len(ino) + 1, 1, masks.dtype)
...