def export_to_bioimageio(checkpoint, output):
input_data = _load_data()
postprocessing = None
offsets = [
[-1, 0], [0, -1],
[-3, 0], [0, -3],
[-9, 0], [0, -9],
[-27, 0], [0, -27]
]
config = {"mws": {"offsets": offsets}}
name = "EpitheliaAffinityModel"
tags = ["u-net", "segmentation"]
cite = get_default_citations(model="UNet2d", model_output="affinities")
doc = "Affinity prediction for epithelia cells"
export_bioimageio_model(
checkpoint, output, input_data,
name=name,
authors=[{"name": "Constantin Pape; @constantinpape"}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
config=config
)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
root, ckpt_name = os.path.split(checkpoint)
organelle = os.path.split(root)[0]
assert organelle in ("cells", "mitochondria", "nuclei"), organelle
if input_ is None:
input_data = None
else:
input_data = _load_data(input_, organelle)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
if organelle in ("cells", ):
postprocessing = "affinities_to_boundaries_anisotropic"
elif organelle in ("mitochondria", "nuclei"):
postprocessing = "affinities_with_foreground_to_boundaries3d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name, description = _get_name_and_description(is_aff_model, organelle)
tags = [
"unet", organelle, "instance-segmentation", "electron-microscopy",
"platynereis"
]
tags += ["boundary-prediction"
] if is_aff_model else ["affinity-prediction"]
cite = get_default_citations(
model="AnisotropicUNet",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://doi.org/10.1101/2020.02.26.961037"
doc = _get_doc(checkpoint, name, organelle)
if additional_formats is None:
additional_formats = []
export_bioimageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
description=description,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("platynereis")])
add_weight_formats(output, additional_formats)
def export_to_bioimageio(checkpoint, input_, output, affs_to_bd,
additional_formats):
ckpt_name = os.path.split(checkpoint)[1]
ndim = 3 if "3d" in ckpt_name else 2
input_data = _load_data(input_, ndim)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = f"affinities_to_boundaries{ndim}d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
if is_aff_model:
offsets = [[-1, 0], [0, -1], [-3, 0], [0, -3], [-9, 0], [0, -9],
[-27, 0], [0, -27]]
config = {"mws": {"offsets": offsets}}
else:
config = {}
name = _get_name(is_aff_model, ndim)
tags = [
"unet", "neurons", "instance-segmentation", "electron-microscopy",
"isbi2012-challenge"
]
tags += ["boundary-prediction"
] if is_aff_model else ["affinity-prediction"]
cite = get_default_citations(
model="UNet2d" if ndim == 2 else "AnisotropicUNet",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://doi.org/10.3389/fnana.2015.00142"
doc = _get_doc(is_aff_model, ndim)
if additional_formats is None:
additional_formats = []
export_biomageio_model(
checkpoint,
output,
input_data,
name=name,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("isbi2012")],
config=config)
add_weight_formats(output, additional_formats)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
ckpt_name = os.path.split(checkpoint)[1]
if input_ is None:
input_data = None
else:
input_data = imageio.imread(input_)
input_data = input_data[:512, :512]
input_data = input_data.transpose((2, 0, 1))
assert input_data.shape[0] == 3, f"{input_data.shape}"
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_with_foreground_to_boundaries2d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name = _get_name(is_aff_model)
tags = ["unet", "instance-segmentation", "nuclei", "whole-slide-imaging"]
tags += ["boundary-prediction"
] if is_aff_model else ["affinity-prediction"]
# eventually we should refactor the citation logic
cite = get_default_citations(
model="UNet2d",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://ieeexplore.ieee.org/document/8880654"
doc = _get_doc(is_aff_model)
if additional_formats is None:
additional_formats = []
export_biomageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
# need custom deepimagej fields if we have torchscript export
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("monuseg")])
add_weight_formats(output, additional_formats)
def export_model():
import imageio
import h5py
from torch_em.util import export_biomageio_model, get_default_citations
from bioimageio.spec.shared import yaml
with h5py.File("./data/gt_image_000.h5", "r") as f:
input_data = f["raw/serum_IgG/s0"][:256, :256]
imageio.imwrite("./cover.jpg", input_data)
doc = "Example Model: Different Output Shape"
cite = get_default_citations(model="UNet2d")
export_biomageio_model(
"./checkpoints/diff-output-shape",
"./exported",
input_data=input_data,
authors=[{"name": "Constantin Pape; @constantinpape"}],
tags=["segmentation"],
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
covers=["./cover.jpg"],
input_optional_parameters=False
)
rdf_path = "./exported/rdf.yaml"
with open(rdf_path, "r") as f:
rdf = yaml.load(f)
# update the shape descriptions
rdf["inputs"][0]["shape"] = {"min": [1, 1, 32, 32], "step": [0, 0, 16, 16]}
rdf["outputs"][0]["shape"] = {"reference_input": "input", "offset": [0, 0, 0, 0], "scale": [1, 1, 0.5, 0.5]}
# update the network description
rdf["source"] = "./resize_unet.py:ResizeUNet"
rdf["kwargs"] = dict(in_channels=1, out_channels=1, depth=3, initial_features=16)
copyfile("./resize_unet.py", "./exported/resize_unet.py")
with open(rdf_path, "w") as f:
yaml.dump(rdf, f)
def export_model():
import h5py
from torch_em.util import export_biomageio_model, get_default_citations
from bioimageio.spec.shared import yaml
with h5py.File("./data/gt_image_000.h5", "r") as f:
input_data = f["raw/serum_IgG/s0"][:256, :256]
doc = "Example Model: Fixed Shape"
cite = get_default_citations(model="UNet2d")
export_biomageio_model(
"./checkpoints/fixed-shape",
"./exported",
input_data=input_data,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=["segmentation"],
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
input_optional_parameters=False)
shape = (1, 1) + input_data.shape
assert len(shape) == 4
# replace the shape
rdf_path = "./exported/rdf.yaml"
with open(rdf_path, "r") as f:
rdf = yaml.load(f)
rdf["inputs"][0]["shape"] = shape
rdf["outputs"][0]["shape"] = shape
with open(rdf_path, "w") as f:
yaml.dump(rdf, f)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
ckpt_name = os.path.split(checkpoint)[1]
if input_ is None:
input_data = None
else:
input_data = _load_data(input_)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_to_boundaries_anisotropic"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name, description = _get_name_and_description(is_aff_model)
tags = [
"unet", "neurons", "instance-segmentation", "electron-microscopy",
"cremi", "connectomics", "3d"
]
tags += ["boundary-prediction"
] if is_aff_model else ["affinity-prediction"]
# eventually we should refactor the citation logic
cite = get_default_citations(
model="AnisotropicUNet",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://cremi.org"
doc = _get_doc(is_aff_model, checkpoint, name)
if is_aff_model:
offsets = [[-1, 0, 0], [0, -1, 0], [0, 0, -1], [-2, 0, 0], [0, -3, 0],
[0, 0, -3], [-3, 0, 0], [0, -9, 0], [0, 0, -9], [-4, 0, 0],
[0, -27, 0], [0, 0, -27]]
config = {"mws": {"offsets": offsets}}
else:
config = {}
if additional_formats is None:
additional_formats = []
export_bioimageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
description=description,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("cremi")],
config=config)
if additional_formats:
add_weight_formats(output, additional_formats)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
ckpt_name = os.path.split(checkpoint)[1]
if input_ is None:
input_data = None
else:
input_data = imageio.imread(input_)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_with_foreground_to_boundaries2d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name, description = _get_name_and_description(is_aff_model)
tags = [
"fluorescence-light-microscopy", "nuclei", "unet",
"instance-segmentation", "2d"
]
# eventually we should refactor the citation logic
cite = get_default_citations(
model="UNet2d",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://www.nature.com/articles/s41592-019-0612-7"
doc = _get_doc(is_aff_model, checkpoint, name)
if is_aff_model:
offsets = [[-1, 0], [0, -1], [-3, 0], [0, -3], [-9, 0], [0, -9],
[-27, 0], [0, -27]]
config = {"mws": {"offsets": offsets}}
else:
config = {}
if additional_formats is None:
additional_formats = []
export_bioimageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
authors=[{
"name": "Constantin Pape",
"affiliation": "EMBL Heidelberg"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
description=description,
# need custom deepimagej fields if we have torchscript export
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("dsb")],
config=config,
maintainers=[{
"github_user": "******"
}],
)
add_weight_formats(output, additional_formats)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
ckpt_name = os.path.split(checkpoint)[1]
if input_ is None:
input_data = None
else:
with h5py.File(input_, "r") as f:
input_data = f["raw/serum_IgG/s0"][:512, :512]
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_with_foreground_to_boundaries2d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name, description = _get_name_and_description(is_aff_model)
tags = [
"unet", "cells", "high-content-microscopy", "instance-segmentation",
"covid19", "immunofluorescence-microscopy", "2d"
]
# eventually we should refactor the citation logic
covid_if_pub = "https://doi.org/10.1002/bies.202000257"
cite = get_default_citations(
model="UNet2d",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = covid_if_pub
if not is_aff_model:
cite["segmentation algorithm"] = covid_if_pub
doc = _get_doc(is_aff_model, checkpoint, name)
if is_aff_model:
offsets = [[-1, 0], [0, -1], [-3, 0], [0, -3], [-9, 0], [0, -9],
[-27, 0], [0, -27]]
config = {"mws": {"offsets": offsets}}
else:
config = {}
if additional_formats is None:
additional_formats = []
export_bioimageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
authors=[{
"name": "Constantin Pape",
"affiliation": "EMBL Heidelberg"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
description=description,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
config=config,
model_postprocessing=postprocessing,
input_optional_parameters=False,
# need custom deepimagej fields if we have torchscript export
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("covid_if")],
maintainers=[{
"github_user": "******"
}],
)
add_weight_formats(output, additional_formats)
def export_model():
import imageio
import h5py
from torch_em.util import add_weight_formats, export_biomageio_model, get_default_citations
from bioimageio.spec.shared import yaml
with h5py.File("./data/gt_image_000.h5", "r") as f:
input_data = [
f["raw/serum_IgG/s0"][:256, :256],
f["raw/nuclei/s0"][:256, :256],
]
imageio.imwrite("./cover.jpg", input_data[0])
doc = "Example Model: Different Output Shape"
cite = get_default_citations(model="UNet2d")
export_biomageio_model(
"./checkpoints/multi-tensor",
"./exported",
input_data=input_data,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=["segmentation"],
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
covers=["./cover.jpg"],
input_optional_parameters=False)
add_weight_formats("./exported", ["onnx", "torchscript"])
rdf_path = "./exported/rdf.yaml"
with open(rdf_path, "r") as f:
rdf = yaml.load(f)
# update the inputs / output descriptions
rdf["inputs"][0]["name"] = "input0"
rdf["inputs"][0]["shape"] = {"min": [1, 1, 32, 32], "step": [0, 0, 16, 16]}
input1 = deepcopy(rdf["inputs"][0])
input1["name"] = "input1"
rdf["inputs"].append(input1)
rdf["outputs"][0]["name"] = "output0"
rdf["outputs"][0]["shape"] = {
"reference_input": "input0",
"offset": [0, 0, 0, 0],
"scale": [1, 1, 1, 1]
}
output1 = deepcopy(rdf["outputs"][0])
output1["name"] = "output1"
output1["shape"]["reference_input"] = "input1"
rdf["outputs"].append(output1)
# update the network description
rdf["source"] = "./multi_tensor_unet.py:MultiTensorUNet"
rdf["kwargs"] = dict(in_channels=2,
out_channels=2,
depth=3,
initial_features=16)
copyfile("./multi_tensor_unet.py", "./exported/multi_tensor_unet.py")
with open(rdf_path, "w") as f:
yaml.dump(rdf, f)
def export_to_bioimageio(checkpoint, output, input_, affs_to_bd,
additional_formats):
root, ckpt_name = os.path.split(checkpoint)
specimen = os.path.split(root)[0]
assert specimen in ("ovules", "roots"), specimen
is2d = checkpoint.endswith('2d')
ndim = 2 if is2d else 3
if input_ is None:
input_data = None
else:
input_data = _load_data(input_, is2d)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_to_boundaries3d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name = _get_name(is_aff_model, specimen, ndim)
tags = [
"u-net", f"{specimen}-segmentation", "segmentation",
"light-microscopy", "arabidopsis"
]
if specimen == "ovules":
tags += ["ovules", "confocal-microscopy"]
else:
tags += ["primordial-root", "light-sheet-microscopy"]
tags += ["boundary-prediction"
] if is_aff_model else ["affinity-prediction"]
# eventually we should refactor the citation logic
plantseg_pub = "https://doi.org/10.7554/eLife.57613.sa2"
cite = get_default_citations(
model="UNet2d" if is2d else "UNet3d",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = plantseg_pub
cite["segmentation algorithm"] = plantseg_pub
doc = _get_doc(is_aff_model, specimen, ndim)
export_biomageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license='CC-BY-4.0',
documentation=doc,
git_repo='https://github.com/constantinpape/torch-em.git',
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
# need custom deepimagej fields if we have torchscript export
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("ovules")])
add_weight_formats(output, additional_formats)
def export_to_bioimageio(checkpoint, input_, output, affs_to_bd,
additional_formats):
root, ckpt_name = os.path.split(checkpoint)
if input_ is None:
input_data = None
else:
input_data = _load_data(input_)
is_aff_model = "affinity" in ckpt_name
if is_aff_model and affs_to_bd:
postprocessing = "affinities_with_foreground_to_boundaries3d"
else:
postprocessing = None
if is_aff_model and affs_to_bd:
is_aff_model = False
name, desc = _get_name_and_description(is_aff_model)
if is_aff_model:
offsets = [[-1, 0, 0], [0, -1, 0], [0, 0, -1], [-2, 0, 0], [0, -3, 0],
[0, 0, -3], [-3, 0, 0], [0, -9, 0], [0, 0, -9]]
config = {"mws": {"offsets": offsets}}
else:
config = {}
cite = get_default_citations(
model="AnisotropicUNet",
model_output="affinities" if is_aff_model else "boundaries")
cite["data"] = "https://doi.org/10.1007/978-3-030-59722-1_7"
tags = [
"3d", "electron-microscopy", "mitochondria", "instance-segmentation",
"unet"
]
doc = _get_doc(is_aff_model, checkpoint, name)
if additional_formats is None:
additional_formats = []
export_bioimageio_model(
checkpoint,
output,
input_data=input_data,
name=name,
description=desc,
authors=[{
"name": "Constantin Pape; @constantinpape"
}],
tags=tags,
license="CC-BY-4.0",
documentation=doc,
git_repo="https://github.com/constantinpape/torch-em.git",
cite=cite,
model_postprocessing=postprocessing,
input_optional_parameters=False,
for_deepimagej="torchscript" in additional_formats,
links=[get_bioimageio_dataset_id("mitoem")],
maintainers=[{
"github_user": "******"
}],
config=config,
)
add_weight_formats(output, additional_formats)