bin2cell.stardist: Found 0 objects

[liangxs7015]

Hi, thanks for develping this tool for us. I have to report a issue during bin2cell.stardist for my 10X HD Visium sample segment. The bin2cell.stardist produced "Found 0 objects" whether I load square_008um or square_002um data
import numpy as np
import matplotlib.pyplot as plt
import scanpy as sc
import tensorflow
import os
import collections
import bin2cell as b2c
import pandas as pd
import gc

#set the path
path = "/home/data/Data_Storage/COAD/n2023262/binned_outputs/square_008um/"
#the image you used for --image of spaceranger, that's the one the spatial coordinates are based on
source_image_path = "/home/data/Data_Storage/Source_image_b2cell/n2023262.jpg"
spaceranger_image_path = "/home/data/Data_Storage/COAD/n2023262/binned_outputs/square_008um/spatial"

#read data
n2023262= b2c.read_visium(path,
source_image_path = source_image_path,
spaceranger_image_path = spaceranger_image_path
)
n2023262.var_names_make_unique()

#filtration
sc.pp.filter_genes(n2023262, min_cells=3)
sc.pp.filter_cells(n2023262, min_counts=1)
n2023262

#scale
mpp = 0.5
b2c.scaled_he_image(n2023262, mpp=mpp, save_path="/home/data/liangxs/R_project/a_Ongoing_SpatialHD/Jupyter/stardist/n2023262_he.tiff")

b2c.destripe(n2023262)

###All the above steps went well
###However, the b2c.stardist produced "Found 0 objects":
b2c.stardist(image_path="/home/data/liangxs/R_project/a_Ongoing_SpatialHD/Jupyter/stardist/n2023262_he.tiff",
labels_npz_path="/home/data/liangxs/R_project/a_Ongoing_SpatialHD/Jupyter/stardist/n2023262_he.npz",
stardist_model="2D_versatile_he",
prob_thresh=0.01
)

_function_base_impl.py (107): overflow encountered in cast
_function_base_impl.py (4750): overflow encountered in cast
_function_base_impl.py (4655): invalid value encountered in multiply
_function_base_impl.py (4656): invalid value encountered in scalar multiply
Found model '2D_versatile_he' for 'StarDist2D'.
Loading network weights from 'weights_best.h5'.
Loading thresholds from 'thresholds.json'.
Using default values: prob_thresh=0.692478, nms_thresh=0.3.
Overriding with prob_thresh=0.01
effective: block_size=(4096, 4096, 3), min_overlap=(128, 128, 0), context=(128, 128, 0)
100%|██████████| 16/16 [01:46<00:00, 6.65s/it]
Found 0 objects

bin2cell version: '0.3.4'

How to solve this problem, please~

[ktpolanski]

_function_base_impl.py (107): overflow encountered in cast
_function_base_impl.py (4750): overflow encountered in cast
_function_base_impl.py (4655): invalid value encountered in multiply
_function_base_impl.py (4656): invalid value encountered in scalar multiply

This part I haven't seen before. I did a little googling and came across this: numpy/numpy#29003

I then tested out the minimal working example they provide with a few different versions of numpy:

import numpy as np
arr = np.zeros(65521, dtype=np.float16)
arr[:10] = 1
z = np.percentile(arr, 50)

The cutoff I found locally is 2.0.0 - the code ran fine on 1.26.4 and blew up on 2.0.0 (and some higher ones I tried). You should have sub-2.x numpy as per the [stardist] specification of the installation, plus from what I remember StarDist just blows up completely if you try to run it with 2.x numpys. What version do you have?

[liangxs7015]

Really aprreciate your quick response! My numpy version is 2.2.6 and I downgraded it to 1.26.4. However, a new report was produced 'module 'keras.backend' has no attribute 'abs''. I'm trying other numpy versions to see if this could solved this issue. Thanks

[ktpolanski]

I don't understand how you ended up in this position, but good that it helped. The new thing you're encountering seems to be tensorflow/stardist related. May I recommend just making a fresh conda environment? I tend to usually use python 3.10, so you can use that one if you want. Then all you'd need to do would be:

pip install bin2cell[stardist]
pip install tensorflow

[liangxs7015]

Yes it seems hard to handle this in current env, I'll try a fresh conda env with python 3.10, thanks a lot

[liangxs7015]

Back here to report that creating an fresh conda env with 'pip install bin2cell[stardist]' and 'pip install tensorflow' finally solved the problem. Really grateful!

[ktpolanski]

As for the initial numpy bug you ran into, it seems that it's on track to be patched in 2.4.0. I'll leave the issue open to act as a reminder for when StarDist updates itself to work with 2.x numpys at some point.