Label transforms ignore include/exclude filters

[Sal-sal-sal]

Description

RemapLabels, RemoveLabels, and OneHot modify label maps that should be excluded by the include or exclude arguments.

Reproduction

import torch
import torchio as tio

subject = tio.Subject(
    a=tio.LabelMap(torch.ones(1, 2, 2, 2)),
    b=tio.LabelMap(torch.ones(1, 2, 2, 2)),
)

result = tio.RemapLabels({1: 9}, include=["a"])(subject)

print(result.a.data.unique())  # tensor([9.])
print(result.b.data.unique())  # tensor([9.]), expected tensor([1.])

The same problem occurs with:

tio.RemapLabels({1: 9}, exclude=["b"])
tio.RemoveLabels([1], include=["a"])
tio.RemoveLabels([1], exclude=["b"])
tio.OneHot(include=["a"])
tio.OneHot(exclude=["b"])

Expected behavior

Only the selected label maps should change.
Excluded label maps should retain their original values and shape.

Possible cause

These transforms iterate over batch.images.items() directly instead of using self._get_images(batch), which applies the include/exclude filters.

Version

Reproduced on main, commit 2b019d2a9659c838408c73ae77327ac5eb87b190, TorchIO 2.0.0a2, using synthetic CPU tensors.

[romainVala]

I confirm the bug, and I am surprise that such a fundamental torchio functionality is missing in this v2 version.
@fepegar how copilot miss this point ?