Closed FynnBe closed 3 years ago
No idea if the model actually excepts the input that I modified in the same manner in https://github.com/deepimagej/models/pull/22 and actually produces this 4d output, so you might want to look into that, too.
This will hopefully fix the failing tests in https://github.com/bioimage-io/spec-bioimage-io/blob/master/tests/test_build_spec.py
Thank you @FynnBe ! I just noticed it today trying Constantin's model. We need to change this in deepImageJ as it saves the example image without considering the batch dimension :(
A wait! could this one depend on the pre-processing?
this particular change is influenced by the post-processing I suppose... test_input -> pre-processing -> nn model -> post-processing -> test_output
But I did change the exampleInput.npy in #22
we really need the dynamic tests, so test_inputs/outputs are actually validated!
added missing batch and channel dimensions to https://github.com/deepimagej/models/blob/master/fru-net_sev_segmentation/resultImage.npy axes of outputs[0] are described as "byxc"