Number of Channels in Images

[vikashg]

Hi all, I got the following error

=== Transform input info -- Resize ===
Data statistics:
Type: <class 'numpy.ndarray'> float32
Shape: (1, 172, 172, 3)
Value range: (0.0, 255.0)
Traceback (most recent call last):
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/MONAI/monai/transforms/transform.py", line 82, in apply_transform
    return _apply_transform(transform, data, unpack_items)
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/MONAI/monai/transforms/transform.py", line 53, in _apply_transform
    return transform(parameters)
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/MONAI/monai/transforms/spatial/array.py", line 442, in __call__
    raise ValueError(
ValueError: len(spatial_size) must be greater or equal to img spatial dimensions, got spatial_size=2 img=3.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/CAII/app_classification/train_device_classifier.py", line 74, in <module>
    _ = train_transforms(image)
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/MONAI/monai/transforms/compose.py", line 160, in __call__
    input_ = apply_transform(_transform, input_, self.map_items, self.unpack_items)
  File "/data/Vikash/Tools_A100/Tools/BreastDensityMonai/MONAI/monai/transforms/transform.py", line 106, in apply_transform
    raise RuntimeError(f"applying transform {transform}") from e
RuntimeError: applying transform <monai.transforms.spatial.array.Resize object at 0x7f09f740d940>

While digging deeper as to the cause of the error and which image might be producing it, I found the images and I found that the spatial dim of the image was -1. All these are .png images and I think the default reader being used is the pilreader.
This is what the image properties of the problematic image looks like

dtype=float32), {'format': 'PNG', 'mode': 'RGB', 'width': 172, 'height': 172, 'spatial_shape': array([172, 172]), **'original_channel_dim': -1, ** 'filename_or_obj': 'file_wrong.png'})

A good image property looks like

dtype=float32), {'format': 'PNG', 'mode': 'L', 'width': 425, 'height': 425, 'spatial_shape': array([425, 425]), **'original_channel_dim': 'no_channel**', 'filename_or_obj': '/filecorrect.png'})

Any thoughts on why the original channel dim is -1 and if we can avoid that.

[Nic-Ma]

Hi @vikashg ,

Sorry I don't quite understand your issue, could you please help share your full test program for further analysis?

Thanks.

[vikashg]

Here is the code that should run

from monai.transforms import (
    Activations,
    LoadImage,
		Resize, ClassesToIndices,
		AsChannelLast, NormalizeIntensity,
		RepeatChannel, RandShiftIntensity,
    ScaleIntensity,
    AddChannel,
    RandRotate,
    EnsureType,
    RandFlip, RandZoom,
    AsDiscrete,
    Compose)


train_transforms = Compose([LoadImage(image_only=True), AddChannel(), Resize((299, 299)), RandShiftIntensity(offsets=20), NormalizeIntensity(), RandFlip(), RandRotate(range_x=2), RepeatChannel(repeats=3), EnsureType()])
train_transforms('./filename.png') 

Here is the sample image
This should reproduce the error.

Thanks a lot @Nic-Ma

[rijobro]

looks like your example image is [172, 172, 3], so AsChannelFirst should be used instead of AddChannel.

In [3]: Compose([LoadImage(image_only=True), AddChannel()])("filename.png").shape
Out[3]: (1, 172, 172, 3)

[rijobro]

If your previous transforms worked for some but not for others, then I suspect that your data consist of some images with the shape [H, W], and others with the shape [H, W, C=3].

[vikashg]

Yes, it might be so. And this is quite common when the data comes directly from scanners without any preprocessing and if we could do something to account for that.

[Nic-Ma]

Hi @vikashg ,

I think you can try to use EnsureChannelFirst transform instead, it can automatically adjust channel dim.

Thanks.