This repository contains the specifications of the standard format defined by the bioimage.io community for the content (i.e., models, datasets and applications) in the bioimage.io website.
Each item in the content is always described using a YAML 1.2 file named rdf.yaml or bioimageio.yaml.
This rdf.yaml \ bioimageio.yaml--- along with the files referenced in it --- can be downloaded from or uploaded to the bioimage.io website and may be produced or consumed by bioimage.io-compatible consumers (e.g., image analysis software like ilastik).
These are the rules and format that bioimage.io-compatible resources must fulfill.
Note that the Python package PyYAML does not support YAML 1.2 . We therefore use and recommend ruyaml. For differences see https://ruamelyaml.readthedocs.io/en/latest/pyyaml.
Please also note that the best way to check whether your rdf.yaml file is bioimage.io-compliant is to call bioimageio.core.validate from the bioimageio.core Python package.
The bioimageio.core Python package also provides the bioimageio command line interface (CLI) with the validate command:
bioimageio validate path/to/your/rdf.yaml
All bioimage.io description formats are defined as Pydantic models.
| Type | Format Version | Documentation1 | Developer Documentation2 |
|---|---|---|---|
| model | 0.5 0.4 |
model 0.5 model 0.4 |
ModelDescr_v0_5 ModelDescr_v0_4 |
| dataset | 0.3 0.2 |
dataset 0.3 dataset 0.2 |
DatasetDescr_v0_3 DatasetDescr_v0_2 |
| notebook | 0.3 0.2 |
notebook 0.3 notebook 0.2 |
NotebookDescr_v0_3 NotebookDescr_v0_2 |
| application | 0.3 0.2 |
application 0.3 application 0.2 |
ApplicationDescr_v0_3 ApplicationDescr_v0_2 |
| generic | 0.3 0.2 |
- | GenericDescr_v0_3 GenericDescr_v0_2 |
Simplified descriptions are available as JSON Schema (generated with Pydantic):
| bioimageio.spec version | JSON Schema | documentation1 |
|---|---|---|
| latest | bioimageio_schema_latest.json | latest documentation |
| 0.5 | bioimageio_schema_v0-5.json | 0.5 documentation |
Note: bioimageio_schema_v0-5.json and bioimageio_schema_latest.json are identical, but bioimageio_schema_latest.json will eventually refer to the future bioimageio_schema_v0-6.json.
A flattened view of the types used by the spec that also shows values constraints.
You can also generate these docs locally by running PYTHONPATH=./scripts python -m interactive_docs
We provide some examples for using rdf.yaml files to describe models, applications, notebooks and datasets, and an example notebook to programmatically access the models, applications, notebooks and datasets descriptions.
- Due to the limitations of storage services such as Zenodo, which does not support subfolders, it is recommended to place other files in the same directory level of the
rdf.yamlfile and try to avoid using subdirectories. - Use the bioimageio.core Python package to validate your
rdf.yamlfile. - bioimageio.spec keeps evolving. Try to use and upgrade to the most current format version!
The bioimageio CLI has moved entirely to bioimageio.core.
bioimageio.spec can be installed with either conda or pip.
We recommend installing bioimageio.core instead to get access to the Python programmatic features available in the BioImage.IO community:
conda install -c conda-forge bioimageio.coreor
pip install -U bioimageio.coreStill, for a lighter package or just testing, you can install the bioimageio.spec package solely:
conda install -c conda-forge bioimageio.specor
pip install -U bioimageio.specTODO: link to settings in dev docs
Made with contrib.rocks.
To keep the bioimageio.spec Python package version in sync with the (model) description format version, bioimageio.spec is versioned as MAJOR.MINRO.PATCH.LIB, where MAJOR.MINRO.PATCH correspond to the latest model description format version implemented and LIB may be bumpbed for library changes that do not affect the format version. This change was introduced with bioimageio.spec 0.5.3.1.
- fix loading tifffile in python 3.8 (pin tifffile)
- use default tensorflow environments for Keras H5 weights
- support loading and saving from/to zipfile.ZipFile objects
- fix bug when packaging with weights priority order (#638)
- add conda_env module providing helper to create recommended conda environments for model descriptions
- fix summary formatting
- improve logged origin for logged messages
- make the
model.v0_5.ModelDescr.training_datafield aleft_to_rightUnion to avoid warnings - the deprecated
version_numberis no longer appended to theid, but instead set asversionif noversionis specified.
- expose
progressbarto customize display of download progress - expose
get_resource_package_content - prefer
rdf.yamloverbioimageio.yaml(namebioimageio.yamlfilerdf.yamlfile when packaging, look forrdf.yamlfirst, etc.) - enforce: (generic 0.3/model 0.5 spec) documentation source file encoding has to be UTF-8.
- bugfix: allow optional pre- and postprocessing to be missing in an RDF (before it required an empty dict).
- bugfix "reset known files if root changes" (#619)
note: the versioning scheme was changed as our previous post releases include changes beyond what a post release should entail (only changing docstrings, etc).
This was motivated by the desire to keep the library version in sync with the (model) format version to avoid confusion.
To keep this relation, but avoid overbearing post releases a library version number is now added as the 4th part MAJOR.MINOR.PATCH.LIB_VERSION.
- add
load_model_descriptionandload_dataset_description - add
ensure_description_is_modelandensure_description_is_dataset - expose
perform_io_checksandknown_filesfromValidationContexttoload_descriptionandload_description_and_validate_format_only
- fix pinning of pydantic
- update resolving of bioimage.io resource IDs
- fix SHA-256 value when resolving a RDF version from the bioimage.io collection that is not the latest
- bump patch version during loading for model 0.5.x
- improve validation error formatting
- validate URLs first with a head request, if forbidden, follow up with a get request that is streamed and if that is also forbidden a regular get request.
RelativePath.absolute()is now a method (not a property) analog topathlib.Path
- remove collection description
- update SPDX license list
- update generic description to 0.3.1
- update model description to 0.5.3
- add timeout argument to all requests.get calls
- added more information to validation summary
- deprioritize
Pathobjects in theFileSourceunion
- resolve backup DOIs
- fix resolving relative file paths given as strings
- allow to bypass download and hashing of known files
- avoid full download when validating urls
- resolve version (un)specific collection IDs, e.g.
load_description('affable-shark'),load_description('affable-shark/1')
- fix model packaging with weights format priority
- new patch version model 0.5.2
- new patch version model 0.5.1
- don't fail if CI env var is a string
- fix
_internal.io_utils.identify_bioimageio_yaml_file()
- new description formats: generic 0.3, application 0.3, collection 0.3, dataset 0.3, notebook 0.3 and model 0.5.
- various API changes, most important functions:
bioimageio.spec.load_description(replacesload_raw_resource_description, interface changed)bioimageio.spec.validate_format(new)bioimageio.spec.dump_description(replacesserialize_raw_resource_description_to_dict, interface changed)bioimageio.spec.update_format(interface changed)
- switch from Marshmallow to Pydantic
- extended validation
- one joint, more precise JSON Schema
- small bugixes
- better type hints
- improved tests
- add
axesandepstoscale_mean_var
- add simple forward compatibility by treating future format versions as latest known (for the respective resource type)
-
Make CLI output more readable
-
find redirected URLs when checking for URL availability
-
Improve error message for non-existing RDF file path given as string
-
Improve documentation for model description's
documentationfield
- fix enrich_partial_rdf_with_imjoy_plugin (see #452)
- fix rdf_update of entries in
resolve_collection_entries()
- pass root to
enrich_partial_rdfarg ofresolve_collection_entries()
- keep
ResourceDescrption.root_pathas URI for remote resources. This fixes the collection description as the collection entries are resolved after the collection description has been loaded.
- new bioimageio.spec.partner module adding validate-partner-collection command if optional 'lxml' dependency is available
-
new env var
BIOIMAGEIO_CACHE_WARNINGS_LIMIT(default: 3) to avoid spam from cache hit warnings -
more robust conversion of ImportableSourceFile for absolute paths to relative paths (don't fail on non-path source file)
- resolve symlinks when transforming absolute to relative paths during serialization; see #438
- fix loading of collection description with id (id used to be ignored)
- support loading bioimageio resources by their animal nickname (currently only models have nicknames).
-
any field previously expecting a local relative path is now also accepting an absolute path
-
load_raw_resource_description returns a raw resource description which has no relative paths (any relative paths are converted to absolute paths).
- add command
commands.update_rdf()/update-rdf(cli)
- fix unresolved ImportableSourceFile
- fix collection description conversion for type field
-
fix to shape validation for model description 0.4: output shape now needs to be bigger than halo
-
moved objects from bioimageio.spec.shared.utils to bioimageio.spec.shared[.node_transformer]
-
additional keys to validation summary: bioimageio_spec_version, status
- fixes to generic description:
- ignore value of field
root_pathif present in yaml. This field is used internally and always present in RDF nodes.
- ignore value of field
- fixes to collection description:
- RDFs specified directly in collection description are validated correctly even if their source field does not point to an RDF.
- nesting of collection description allowed
-
fixed missing field
iconin generic description's raw node -
fixes to collection description:
- RDFs specified directly in collection description are validated correctly
- no nesting of collection description allowed for now
linksis no longer an explicit collection entry field ("moved" to unknown)
- new model spec 0.3.5 and 0.4.1
load_raw_resource_descriptionno longer acceptsupdate_to_current_formatkwarg (useupdate_to_formatinstead)
load_raw_resource_descriptionacceptsupdate_to_formatkwarg
- Non-breaking changes
- remove
version_numberin favor of usingversion
- remove
- Non-breaking changes
- added
concatenableflag to index, time and space input axes
- added
- Non-breaking changes
- added
DataDependentSizeforoutputs.i.sizeto specify an output shape that is not known before inference is run. - added optional
inputs.i.optionalfield to indicate that a tensor may beNone - made data type assumptions in
preprocessingandpostprocessingexplicit by adding'ensure_dtype'operations per default. - allow to specify multiple thresholds (along an
axis) in a 'binarize' processing step
- added
- Breaking canges that are fully auto-convertible
- dropped
download_url - dropped non-file attachments
attachments.filesmoved toattachments.i.source
- dropped
- Non-breaking changes
- added optional
parentfield
- added optional
all generic 0.3.0 changes (except models already have the parent field) plus:
- Breaking changes that are partially auto-convertible
inputs.i.axesare now defined in more detail (same foroutputs.i.axes)inputs.i.shapemoved per axes toinputs.i.axes.size(same foroutputs.i.shape)- new pre-/postprocessing 'fixed_zero_mean_unit_variance' separated from 'zero_mean_unit_variance', where
mode=fixedis no longer valid. (for scalar values this is auto-convertible.)
- Breaking changes that are fully auto-convertible
- changes in
weights.pytorch_state_dict.architecture- renamed
weights.pytorch_state_dict.architecture.source_fileto...architecture.source
- renamed
- changes in
weights.pytorch_state_dict.dependencies- only conda environment allowed and specified by
weights.pytorch_state_dict.dependencies.source - new optional field
weights.pytorch_state_dict.dependencies.sha256
- only conda environment allowed and specified by
- changes in
weights.tensorflow_model_bundle.dependencies- same as changes in
weights.pytorch_state_dict.dependencies
- same as changes in
- moved
test_inputstoinputs.i.test_tensor - moved
test_outputstooutputs.i.test_tensor - moved
sample_inputstoinputs.i.sample_tensor - moved
sample_outputstooutputs.i.sample_tensor - renamed
inputs.i.nametoinputs.i.id - renamed
outputs.i.nametooutputs.i.id - renamed
inputs.i.preprocessing.nametoinputs.i.preprocessing.id - renamed
outputs.i.postprocessing.nametooutputs.i.postprocessing.id
- changes in
- Non-breaking changes:
- new pre-/postprocessing:
id='ensure_dtype' with kwargdtype
- new pre-/postprocessing:
- Breaking changes that are fully auto-convertible
idoverwritten with value fromconfig.bioimageio.nicknameif available
- Non-breaking changes
version_numberis a new, optional field indicating that an RDF is the nth published version with a givenidid_emojiis a new, optional field (set fromconfig.bioimageio.nickname_iconif available)uploaderis a new, optional field withemailand an optionalnamesubfields
- Non-breaking changes
- make pre-/postprocessing kwargs
modeandaxesalways optional for model description 0.3 and 0.4
- make pre-/postprocessing kwargs
- Non-breaking changes
citefield is now optional
- Breaking changes that are fully auto-convertible
- name field may not include '/' or '' (conversion removes these)
- Non-breaking changes
- Implicit output shape can be expanded by inserting
nullintoshape:scaleand indicating length of new dimension D in theoffsetfield. Keep in mind thatD=2*'offset'.
- Implicit output shape can be expanded by inserting
- Breaking changes that are fully auto-convertible
parentfield changed to hold a string that is a bioimage.io ID, a URL or a local relative path (and not subfieldsuriandsha256)
- Non-breaking changes
- new optional field
training_data
- new optional field
- Non-breaking changes
- explicitly define and document dataset description (for now, clone of generic description with type="dataset")
- Non-breaking changes
- add optional field
download_url - add optional field
dependenciesto all weight formats (not only pytorch_state_dict) - add optional
pytorch_versionto the pytorch_state_dict and torchscript weight formats
- add optional field
- Bug fixes:
- in a
pytorch_state_dictweight entryarchitectureis no longer optional.
- in a
-
Non-breaking changes
- make
authors,cite,documentationandtagsoptional
- make
-
Breaking changes that are fully auto-convertible
- Simplifies collection description 0.2.1 by merging resource type fields together to a
collectionfield, holindg a list of all resources in the specified collection.
- Simplifies collection description 0.2.1 by merging resource type fields together to a
- Non-breaking changes
rdf_sourcenew optional fieldidnew optional field
- First official release, extends generic description with fields
application,model,dataset,notebookand (nested)collection, which hold lists linking to respective resources.
- Non-breaking changes
- add optional
emailandgithub_userfields to entries inauthors - add optional
maintainersfield (entries like inauthorsbutgithub_useris required (andnameis not))
- add optional
-
Breaking changes that are fully auto-convertible
- moved field
dependenciestoweights:pytorch_state_dict:dependencies
- moved field
-
Non-breaking changes
documentationfield accepts URLs as well
- Non-breaking changes
documentationfield accepts URLs as well
-
Breaking changes
- model inputs and outputs may not use duplicated names.
- model field
sha256is required ifpytorch_state_dictweights are defined. and is now moved to thepytroch_state_dictentry asarchitecture_sha256.
-
Breaking changes that are fully auto-convertible
- model fields language and framework are removed.
- model field
sourceis renamedarchitectureand is moved together withkwargsto thepytorch_state_dictweights entry (if it exists, otherwise they are removed). - the weight format
pytorch_scriptwas renamed totorchscript.
-
Other changes
- model inputs (like outputs) may be defined by
scaleing andoffseting areference_tensor - a
maintainersfield was added to the model description. - the entries in the
authorsfield may now additionally containemailorgithub_user. - the summary returned by the
validatecommand now also contains a list of warnings. - an
update_formatcommand was added to aid with updating older RDFs by applying auto-conversion.
- model inputs (like outputs) may be defined by
- Non-breaking changes
- Add optional parameter
epstoscale_rangepostprocessing.
- Add optional parameter
- Breaking changes that are fully auto-convertible
reference_inputfor implicit output tensor shape was renamed toreference_tensor
-
Breaking changes
- The RDF file name in a package should be
rdf.yamlfor all the RDF (notmodel.yaml); - Change
authorsandpackaged_byfields from List[str] to List[Author] with Author consisting of a dictionary{name: '<Full name>', affiliation: '<Affiliation>', orcid: 'optional orcid id'}; - Add a mandatory
typefield to comply with the generic description. Only valid value is 'model' for model description; - Only allow
licenseidentifier from the SPDX license list;
- The RDF file name in a package should be
-
Non-breaking changes
- Add optional
versionfield (default 0.1.0) to keep track of model changes; - Allow the values in the
attachmentslist to be any values besides URI;
- Add optional
Footnotes
-
JSON Schema based documentation generated with json-schema-for-humans. ↩ ↩2
-
Part of the bioimageio.spec package documentation generated with pdoc. ↩