FAST with FastPathology pipeline, export with fast.ImageExporter

[mhoibo]

Hi, I have tried to use a pipeline from the fast pathology datahub (breast epithelium segmentation) with png-images from a public dataset that are not pyramidal. I edited the pipeline instead of using it out of the box, and used ImageImporter and ImageExporter:

model = model_onnx = '/home/user/FAST/datahub/breast-epithelium-segmentation-model/' \
             'breast_epithelium_segmentation_single_output.onnx'

importer = fast.ImageImporter.create(path_)  # path to png image

tissue_seg = fast.TissueSegmentation.create(threshold=70).connect(importer)

generator = fast.PatchGenerator.create(1024, 1024, overlapPercent=0.3,
                                       maskThreshold=0.1).connect(0, importer).connect(1, tissue_seg)

network = fast.SegmentationNetwork.create(model_, scaleFactor=1./255.). \
    connect(generator)

stitcher = fast.PatchStitcher.create().connect(network)

finished = fast.RunUntilFinished.create().connect(stitcher)

# save segmentation on disk
os.makedirs(seg_path, exist_ok=True)
fast.ImageExporter.create(file_path) \
    .connect(finished).run()

But i get an error on the ImageExporter():

 RuntimeError: Bad cast. Tried to cast from type Image to ImagePyramid

Do you know why the export is problematic, and if there is a possible solution to it? I have used inputs to this model before which were not vsis, but images made with fast.Image.createFromArray(). If I convert to numpy, then create a fast image with it I get the exact same error. Other exporters also yield the same error.

[smistad]

Hi. I believe this error is coming from TissueSegmentation which expects an ImagePyramid. You can verify that by removing it.

I will add a small fix to TissueSegmentation which makes it support regular images as well.

[mhoibo]

Hi, yes, that worked. Thank you, @smistad

[smistad]

Here is a fix for TissueSegmentation ca375a6 @mhoibo
Now you can send a regular Image object to TissueSegmentation.