Sometimes, even when we set mask_circle=True in function squidpy.im.calculate_image_features, the features of segmentation is still calculated by a square mask rather than a circle mask.
After using sq.im.calculate_image_features, if we check the distance between each spot center and the corresponding centroids, the maximum distance is larger than spot radius and less than sqrt(2)*spot radius. By using the following code, we can also easily tell that the mask is still square, not circle.
df = adata_st.obsm["image_features"]['segmentation_centroid']
spot_center = adata_st.obsm['spatial']
d_max = np.zeros(len(spot_center))
diff= []
for i in tqdm(range(len(spot_center))):
d = 0
for location in df[i]:
if len(location) > 0 and location[0] is not np.nan:
diff += [np.array(location)-spot_center[i]]
d = max(d, np.sqrt(np.sum((diff[-1])**2)))
d_max[i] = d
print(max(d_max)) # 52.3
diff = np.array(diff)
mpl.pyplot.scatter(diff[:, 0], diff[:, 1], s=1)
mpl.pyplot.axis('equal')
Another piece of evidence is the very last figure of this tutorial. If the mask is correct, then the centers of all dots should be in spots.
Description
Sometimes, even when we set
mask_circle=True
in functionsquidpy.im.calculate_image_features
, the features of segmentation is still calculated by a square mask rather than a circle mask.Minimal reproducible example
The simplest reproducible example is this tutorial: https://squidpy.readthedocs.io/en/stable/notebooks/tutorials/tutorial_tangram.html
After using
sq.im.calculate_image_features
, if we check the distance between each spot center and the corresponding centroids, the maximum distance is larger than spot radius and less than sqrt(2)*spot radius. By using the following code, we can also easily tell that the mask is still square, not circle.Another piece of evidence is the very last figure of this tutorial. If the mask is correct, then the centers of all dots should be in spots.