cv2.imread（）读取的图文维度存在随机性

Expected behaviour

Write here how did you expect the library to function.

Actual behaviour

(160, 256)
(160, 256, 1)
(160, 256)
(160, 256, 1)

和

(160, 256)
(160, 256, 1)
(160, 256, 1)
(160, 256, 1)

多次运行，会不断改变，我不太明白是什么原因。我希望能保持单通道的1这个参数，在我的结果里。

Steps to reproduce

fundus_img = cv2.imread('/root/autodl-fs/Imag_split/processed_data/1-1 dose/0_0.jpg', cv2.IMREAD_GRAYSCALE)
print(fundus_img.shape)
name = np.expand_dims(fundus_img, axis=2)
print(name.shape)
nameimg = name.astype(np.float32)
nameimg2 = img_train_transforms(nameimg)
print(nameimg2.shape)
if len(nameimg2.shape) != 3:
nameimg2 = np.expand_dims(nameimg2, axis=2)

print(nameimg2.shape)

• operating system
  linux
• architecture (e.g. x86)
  x86
• opencv-python version：opencv-python 4.10.0.84

Issue submission checklist

• This is not a generic OpenCV usage question (looking for help for coding, other usage questions, homework etc.)
• I have read the README of this repository and understand that this repository provides only an automated build toolchain for OpenCV Python packages (there is no actual OpenCV code here)
• The issue is related to the build scripts in this repository, to the pre-built binaries or is a feature request (such as "please enable this additional dependency")
• I'm using the latest version of opencv-python

Start by running the supplied cv2.imread and img_train_transforms snippet, checking the shapes before and after each step. No repository file or test is identified, and the issue concerns general OpenCV usage rather than this build toolchain; a useful resolution would first isolate which operation changes the channel dimension.