[BUG] possibly issues with anisotropy in cellpose/gui/make_train.py

[MariaTheiss]

Describe the bug
I am running make_train on a single-channel tif image with dimensions (z, y, x). I set --anisotropy = 3. When opening the resulting crops in the Cellpose GUI, anisotropy is applied to all yx, zx, and zy crops. I expected corrections for anisotropy only along the z-axis.

To Reproduce
Steps to reproduce the behavior:

1. To check image dimensions in python:

import tifffile as tiff

# Load the TIFF file
folder = 'folder/to/image/'
file = 'example_image.tif'

file_path = folder + file
image = tiff.imread(file_path)
# Check the image shape
print(f"Image dimensions: {image.shape}")

Output:
Image dimensions: (23, 2304, 6452) (This corresponds to z, y, x)

2. Run make_train: python make_train.py --dir directory/to/the/image --z_axis 0 --anisotropy 3
3. To open cellpose GUI: python -m cellpose --Zstack
4. When dragging and dropping the resulting images from directory/to/the/image/train into the cellpose GUI, all crops (YX, ZX, and ZY) are stretched along one axis.

Attached: showing anisotropy in yx (nuclei appear stretched).

[JellyfishKz]

Hello, I have also encountered a similar problem. My problem is that during the recognition process of a 3D grayscale image with a shape of (2304 * 2304 * 10), the z-axis, x-axis, and y-axis of the grayscale image are always misread due to the logic of "If channel_axis=None cellpose will try to automatically determine the channel axis by choosing the dimension with the minimum size after squeezing. I think your problem may be similar to mine

[MariaTheiss]

Hello, I have also encountered a similar problem. My problem is that during the recognition process of a 3D grayscale image with a shape of (2304 * 2304 * 10), the z-axis, x-axis, and y-axis of the grayscale image are always misread due to the logic of "If channel_axis=None cellpose will try to automatically determine the channel axis by choosing the dimension with the minimum size after squeezing. I think your problem may be similar to mine

Thank you! Did you manage to find a workaround?

[JellyfishKz]

您的作业已收到，谢谢