Saving 3D Tensor to Multi-page TIFF using SaveImage

[muell135]

I'm in the process of equipping my ML pipeline to accept 3d TIFFs as training data, but I'm running into problems with saving the patches. It seems TIFF doesn't support 32-bit 3d images, but the SaveImage call func does not allow me to specify a different output_dtype for the PILWriter:

writer_obj = writer_cls(**self.init_kwargs)

I tried passing init_kwargs={'output_dtype': np.int8} to SaveImage.set_options , but it it doesn't seem to care since output_dtype has a default value of np.float32 and doesn't look in kwargs for this arguement.

Is it even possible to use SaveImage to save 3d TIFFs then?

Thanks!

[KumoLiu]

Hi @muell135, to save the image in PNG format, you can set the output_dtype to np.uint8, np.uint16, or None (which will default to the input data's dtype). You can find more details in the code here:
MONAI Array Transform Code.

For saving the data in TIFF format, you'll need to specify the output_ext accordingly.

Hope it helps, thanks.

[muell135]

This is my test script:

import numpy as np
import torch
from monai.data import MetaTensor
from monai.transforms import Compose, SaveImage
from pathlib import Path


# Create a random 3x3x3 tensor (with an added channel dimension)
tensor = torch.randint(1, 3, (1, 3, 3, 3))  # random integers between 1 and 2 (upper bound is exclusive)
meta_tensor = MetaTensor(tensor)

# Define the save path
save_path = Path(r'...')


# not sure if i'm doing this correctly
init_kwargs={'output_dtype': np.uint8}
data_kwargs={'spatial_ndim': 3}
meta_kwargs={}
write_kwargs={'format':'TIFF', 'save_all': True, 'compression':"tiff_lzw", 'output_dtype': np.uint8, 'image_mode': 'L'}  


save1 = SaveImage(output_dir=save_path, # Directory where TIFF files will be saved
                   output_ext=".tiff",  # File extension, use ".tiff" for individual TIFF files
                   output_postfix="",  # Add a postfix to filenames if desired (e.g., '_seg')
                   resample=False,  # Disable resampling if unnecessary
                   separate_folder=False,  # Save files in the output_dir directly, not in subfolders
                   print_log=False,
                  squeeze_end_dims=False)
save1.set_options(init_kwargs=init_kwargs, data_kwargs=data_kwargs, write_kwargs=write_kwargs)
save1(meta_tensor)

but I keep getting:

Traceback (most recent call last):
  File "C:\Program Files\JetBrains\PyCharm Community Edition 2024.2\plugins\python-ce\helpers\pydev\pydevd.py", line 1557, in _exec
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "C:\Program Files\JetBrains\PyCharm Community Edition 2024.2\plugins\python-ce\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "S:\PyCharm_Projects\Project\01\test2.py", line 40, in <module>
    writer_obj.write(save_path, **write_kwargs)
  File "C:\Users\user\.conda\envs\env\lib\site-packages\monai\data\image_writer.py", line 774, in write
    self.data_obj = self.create_backend_obj(
  File "C:\Users\user\.conda\envs\env\lib\site-packages\monai\data\image_writer.py", line 862, in create_backend_obj
    return PILImage.fromarray(data, mode=kwargs.pop("image_mode", None))
  File "C:\Users\user\.conda\envs\env\lib\site-packages\PIL\Image.py", line 3113, in fromarray
    raise ValueError(msg)
ValueError: Too many dimensions: 3 > 2.