为单张缩放图片添加AdditiveGaussianNoise遇AssertionError求代码修正

Fixing the AssertionError & Image Range Issue #

First, let's address the immediate error you're seeing:

The AssertionError occurs because when you call aug(resized_img), the augmenter interprets your image array as the return_batch argument (which expects a boolean value, not a numpy array). Imgaug requires you to explicitly pass images using named arguments or wrap single images in a list (since it's designed to handle batches of images).

Solution 1: Use the images keyword parameter

This is the most straightforward fix:

augmented_image = aug(images=resized_img)

Solution 2: Wrap the image in a list

If you prefer this syntax, just remember to extract the first element (since it returns a batch list):

augmented_image = aug([resized_img])[0]

Second Critical Fix: Image Intensity Range #

You'll run into a second issue even after fixing the error: skimage.transform.resize returns images with pixel values in the [0, 1] range (as floats), but your AdditiveGaussianNoise scale is set to (0, 0.2*255) (adding noise in the 0-51 range). This will push pixel values way outside the valid [0,1] range, causing artifacts or downstream errors.

You have two options to resolve this:

Option A: Convert to 0-255 uint8 format

Scale the resized image back to the standard 8-bit range before applying noise:

import numpy as np  # Don't forget to import numpy

resized_img = resize(imread(file_name), (224, 224)) * 255
resized_img = resized_img.astype(np.uint8)

Option B: Adjust noise scale for [0,1] range

Keep the image in float format and reduce the noise scale to match:

aug = iaa.AdditiveGaussianNoise(scale=(0, 0.2))

Full Corrected Code Example #

Here's the complete code using the [0,1] range approach:

from skimage.io import imread
from skimage.transform import resize
import imgaug.augmenters as iaa

file_name = "path/to/image.jpg"
resized_img = resize(imread(file_name), (224, 224))

# Adjust noise scale to match [0,1] image range
aug = iaa.AdditiveGaussianNoise(scale=(0, 0.2))
# Use named argument to avoid AssertionError
augmented_image = aug(images=resized_img)

内容的提问来源于stack exchange，提问作者Tobitor