Blend image-label to check the segmentation region

[HripsimeS]

Dear MONAI's team @wyli @Nic-Ma @yiheng-wang-nv

I am trying to optimize your method to blend the image and label to check the segmentation region (cells 11-12).
https://github.com/Project-MONAI/tutorials/blob/master/modules/transform_visualization.ipynb

Here is my notebook, please take a look https://github.com/HripsimeS/Computer-Vision/blob/main/3DUnet_GitHub..ipynb

It worked with image and true mask (cells 10-11) as they had the same shape (single channel), although on the 3rd column blend of all labels have the same red color (I have 4 labels excluding the background).
But when I am trying to do blend of image and predicted mask (cells 30-31), I get an error because my predicted mask has 5 channels and image has a single channel. Can you please help to figure out how to fix this channel issue and how to have different colors of each labels on blend images for the 3rd column. Look forward to hearing from you!

[rijobro]

Strange. Can paste the results of check_data["mask"][0].dtype and torch.unique(check_data["mask"][0])? Also check_data["image"].max()

[HripsimeS]

@rijobro thanks a lot, now my label shape is torch.Size([1, 512, 512, 313]). But now got other error "Image should have 1 or 3 channels" because image shape is (512, 512, 313).
I assume the error comes from this code original_image = loader(test_data[0]["image_meta_dict"]["filename_or_obj"])[0]
When I remove [0] near test_data, I receive "list indices must be integers or slices, not str"
When I remove [0] from the end, I receive the following "'tuple' object has no attribute 'shap"

[rijobro]

Add channel component to your original image:

rety = blend_images(image=original_image[None], label=test_output_argmax, alpha=0.5, cmap="hsv", rescale_arrays=True)

[rijobro]

You mix [B,C,H,W,D], [C,H,W,D] and [H,W,D] throughout your code. It would be best to be clear about the shape required at each step to avoid these problems. Also there's no need to re-load your image (original_image = loader(test_data[0]["image_meta_dict"]["filename_or_obj"])[0]) as it is already image_data["image"] in a different shape.

[HripsimeS]

@rijobro thanks a lot, eventually it worked out 👍 🥇

[rijobro]

Looks great!