Scipy.misc无imresize/imread属性，求适配Google Colab的替代代码

Hey there! It’s super common to hit this deprecated scipy issue these days—those old scipy.misc image functions were retired a while back, so Google Colab’s up-to-date scipy install doesn’t include them anymore. Let’s swap them out for the PIL-based approach you mentioned, step by step.

Step 1: Replace scipy.misc.imread #

First, we’ll replace sc.imread() with PIL’s Image.open(), which is the standard modern way to load images. We’ll also ensure the image is in RGB mode to match your original mode='RGB' parameter.

Step 2: Replace scipy.misc.imresize #

For resizing, we’ll use PIL’s built-in resize() method with bilinear interpolation, then convert back to a numpy array and cast to double precision just like your original code.

Full Modified Code #

Here’s the drop-in replacement for your original snippet:

from PIL import Image
import numpy as np

height = 256 
width = 256 
channels = 3 

# Load the image with PIL instead of scipy.misc.imread
img = Image.open(Tr_list[idx])
# Ensure image is in RGB mode (matches your original mode='RGB')
img = img.convert('RGB')
# Resize with bilinear interpolation (note: PIL uses (width, height) order!)
img_resized = img.resize((width, height), resample=Image.Resampling.BILINEAR)
# Convert to numpy array and cast to double precision
img = np.double(img_resized)

Key Notes to Avoid Gotchas: #

• Dimension Order: Unlike scipy.imresize which uses [height, width], PIL’s resize() takes (width, height)—don’t mix these up, or your image will be stretched incorrectly!
• Interpolation: Image.Resampling.BILINEAR is the modern PIL syntax (for versions ≥9.1.0). If you’re on an older PIL version, use Image.BILINEAR instead—Colab’s current PIL should support the Resampling enum, but it’s good to keep this in mind.
• RGB Conversion: The convert('RGB') step ensures even grayscale or RGBA images get converted to 3-channel RGB, matching your original mode='RGB' requirement.

This code will behave exactly like your original snippet but uses active, supported libraries instead of deprecated scipy functions.

内容的提问来源于stack exchange，提问作者Babak Azad