Scaling of RandImageFilterd transform

[sebquetin]

Describe the bug
When using monai.transforms.RandImageFilterd, the scale/distribution of the output of the transform changes from the input one.

To Reproduce
Steps to reproduce the behavior:
"""
import monai
import torch

print("I am using monai ", monai.version)
print("I am using torch ", torch.version)

transform = monai.transforms.Compose([monai.transforms.RandImageFilterd(keys="image", kernel="mean",kernel_size=3, prob=1)])

inpt = torch.randint(low=0, high=255, size=(224, 224))
outpt = transform({"image": inpt})

print(inpt.max())
print(inpt.min())
print(outpt["image"].max())
print(outpt["image"].min())
"""

Expected behavior
I would expect the max value of the transformed image to be lower than the max value of the input since we are applying a mean filter to smooth edges of the image. Should the output of the filter be divided by kernel_size2 for an image and kernel_size3 for a volume?
I observe the same problem with other filters. Is it expected for the user to create a lambda transform after the RandImageFilterd transform to scale the image back properly?

Screenshots

Environment

Printing MONAI config...

MONAI version: 1.2.0
Numpy version: 1.25.1
Pytorch version: 2.0.1+cpu
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: c33f1ba
MONAI file: /home/sebquet/.local/lib/python3.10/site-packages/monai/init.py

Optional dependencies:
Pytorch Ignite version: NOT INSTALLED or UNKNOWN VERSION.
ITK version: NOT INSTALLED or UNKNOWN VERSION.
Nibabel version: NOT INSTALLED or UNKNOWN VERSION.
scikit-image version: 0.21.0
Pillow version: 9.0.1
Tensorboard version: 2.13.0
gdown version: NOT INSTALLED or UNKNOWN VERSION.
TorchVision version: 0.15.2+cpu
tqdm version: 4.65.0
lmdb version: NOT INSTALLED or UNKNOWN VERSION.
psutil version: NOT INSTALLED or UNKNOWN VERSION.
pandas version: 2.0.3
einops version: NOT INSTALLED or UNKNOWN VERSION.
transformers version: NOT INSTALLED or UNKNOWN VERSION.
mlflow version: NOT INSTALLED or UNKNOWN VERSION.
pynrrd version: 1.0.0

For details about installing the optional dependencies, please visit:
https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies

================================
Printing system config...

psutil required for print_system_info

================================
Printing GPU config...

Num GPUs: 0
Has CUDA: False
cuDNN enabled: False

[KumoLiu]

Hi @sebquetin, good points.
Seems we need to either improve our implementation or update the docstring to instruct the user to manually scale the image.

[vikashg]

@KumoLiu can you update the docstring.

[K-Rilla]

Our team from CodeDayLabs '24 will be working on this issue.