Description:

I'm running a deep learning pipeline that relies on transform.resize, but when uint16 images are passed to it, it returns them as float64 (with the same value range, not rescaled). This causes downstream issues due to many packages not supporting float64. Why would transform.resize change the dtype?

Way to reproduce:

import numpy as np
from skimage.transform import resize
img = np.random.uniform(0,65535,10000)
img = img.reshape(100,100)
img = img.astype(np.uint16)
resized = resize(img, (50, 50), preserve_range=True, anti_aliasing=True)
print(f"original dtype: {img.dtype} | resized dtype: {resized.dtype}")

Version information:

3.10.14 (main, May  6 2024, 19:42:50) [GCC 11.2.0]
Linux-6.9.12-200.fc40.x86_64-x86_64-with-glibc2.39
scikit-image version: 0.24.0
numpy version: 1.26.4