Is your feature request related to a problem? Please describe.
I’m using distributed_segmentation on large 2D multi-channel images with shape (C, H, W). To make the Zarr array tile cleanly, the blocksize must be 3D (e.g. (3, 256, 256)), where 3 is the number of channels. Downstream utilities appear to treat the first axis as a spatial axis rather than a channel axis. For example:
remove_overlaps reduces the result to (3, 256) instead of the expected spatial shape (256, 256).- The output/empty Zarr created for labels inherits the 3D chunking (3, 256, 256) from the input, even though the segmentation output is 2D per image (H, W) and should be chunked like (256, 256).
It’s unclear whether multi-channel 2D inputs are unsupported, or if I’ve misconfigured parameters.
Describe the solution you'd like
Should there be an additional parameter to configure the channel axis for distributed_segmentation to treat specified axis as channels and correctly handle them?
Is your feature request related to a problem? Please describe.
I’m using
distributed_segmentationon large 2D multi-channel images with shape (C, H, W). To make the Zarr array tile cleanly, the blocksize must be 3D (e.g. (3, 256, 256)), where 3 is the number of channels. Downstream utilities appear to treat the first axis as a spatial axis rather than a channel axis. For example:remove_overlapsreduces the result to (3, 256) instead of the expected spatial shape (256, 256).It’s unclear whether multi-channel 2D inputs are unsupported, or if I’ve misconfigured parameters.
Describe the solution you'd like
Should there be an additional parameter to configure the channel axis for
distributed_segmentationto treat specified axis as channels and correctly handle them?