[BUG] running model with compute_masks=False causes cv2.resize error because masks=[0]

[fyz11]

I just upgraded to cellpose 4.0.4 today.

I ran the pretrained 'cyto' model with compute_masks=False, and a diameter value which was not None. This led to a cv2.resize error.

Tracing the error it occurs in models.py with the model.eval. When compute_masks=False, masks=[0]. However, it is the current code is still trying to resize this back to the input image resolution. Please could you fix this? I have drawn an arrow to indicate the line needing to fix - thanks !

[mrariden]

Hi @fyz11 Please post a complete description of the code you ran to produce this error. Are you using the GUI, and if so what settings? If you are using the CLI/API please post a minimal code example that produces the error. What is the shape of the input image? Also please post both the output of the welcome message (the python version, and OS) and the output of pip list in the environment you are running cellpose from.

[fyz11]

Sure thing.

I am running Python 3.9.21 in a conda environment, and running cellpose through the code API.

This is the screenshot of the welcome message.

This is a minimal code snippet to reproduce the error.

  import skimage.io as skio 
  import pylab as plt 
  import numpy as np 
  import scipy.ndimage as ndimage
  import skimage.segmentation as sksegmentation 
  import os 
  
  import cellpose
  import cellpose.models as models
  
  
  """
  Load the image
  """
  trainfolder = '/archive/bioinformatics/shared/U54_segment/Public_Datasets_for_CellSegmentation/Cellpose_Dataset/Training/train'
  
  imgfile = os.path.join(trainfolder,
                         '000_img.png')
  img = skio.imread(imgfile)
  
  
  
  """
  This breaks, when diameter=20 and compute_masks=False
  """
  cp_model = models.CellposeModel(gpu=True, pretrained_model='cyto')
  masks, flow, style = cp_model.eval(img,
                                     diameter=60, 
                                     channels=[0,0],
                                     cellprob_threshold=-np.inf,
                                     flow_threshold=0.85, 
                                     compute_masks=False)
  _, flow, prob = flow
  
  plt.figure()
  plt.imshow(prob)
  plt.show()

The image I am using is the train/000_img.png from the cellpose training dataset for illustration (downloaded from https://www.[cellpose](https://www.cellpose.org/).org/). The error doesn't matter what the dimensions are. The trigger conditions is a custom value for diameter and compute_masks=False.

I attach my pip list output.
cellpose4_error_env.txt

Thanks for looking into this.