Parser and validator library for BioImage.IO specifications
Project description
Specifications for BioImage.IO
This repository contains specifications defined by the BioImage.IO community. These specifications are used for defining fields in YAML files which we called Resource Description Files
or RDF
. The RDFs can be downloaded or uploaded to the bioimage.io website, produced or consumed by BioImage.IO-compatible consumers(e.g. image analysis software or other website). Currently we defined two types of RDFs: a dedicated RDF specification for AI models (i.e. model RDF
) and a general RDF specification. The model RDF is a RDF with additional fields that specifically designed for describing AI models.
All the BioImage.IO-compatible RDF must fulfill the following rules:
- Must be a YAML file encoded as UTF-8; If yaml syntax version is not specified to be 1.1 in the first line by
% YAML 1.1
it must be equivalent in yaml 1.1 and yaml 1.2. For differences see https://yaml.readthedocs.io/en/latest/pyyaml.html#differences-with-pyyaml. - The RDF file extension must be
.yaml
(not.yml
) - The RDF file can be saved in a folder (or virtual folder) or in a zip package, the following additional rules must apply:
- When stored in a local file system folder, github repo, zenodo deposition, blob storage virtual folder or similar kind, the RDF file name should match the pattern of
*.rdf.yaml
, for examplemy-model.rdf.yaml
. - When the RDF file and other files are zipped into a RDF package, it must be named as
rdf.yaml
.
- When stored in a local file system folder, github repo, zenodo deposition, blob storage virtual folder or similar kind, the RDF file name should match the pattern of
As a general guideline, please follow the model RDF spec to describe AI models and use the general RDF spec for other resource types including dataset
, notebook
, application
. You will find more details about these two specifications in the following sections. Please also note that the best way to check whether your RDF file is BioImage.IO-compliant is to run the BioImage.IO Validator against it.
Resource Description File Specification
A BioImage.IO-compatible Resource Description File (RDF) is a YAML file with a set of specifically defined fields.
You can find detailed field definitions here:
The specifications are also available as json schemas:
Describing applications
The RDF can be used to describe applications. To do so set the type
field to application
.
For regular software package with a downloadable file, you can set download_url
to the downloadable file, for example, you can upload the executable files as Github release, deposit it on Zenodo, or even generate a sharable url from Dropbox/Google Drive.
For web application, set source
to the url of the web application. Users can then click and redirect to your web application. However, simple integration will not support features such as opening dataset or models with your application.
It is recommended to build BioEngine Apps such that users can directly try and use them in BioImage.IO. See here for more details.
Below is an example for Kaibu, which is a BioEngine/ImJoy compatible web application:
id: kaibu
name: Kaibu
description: Kaibu--a web application for visualizing and annotating multi-dimensional images
covers:
# use the `raw` url if you store the image on github
- https://raw.githubusercontent.com/imjoy-team/kaibu/master/public/static/img/kaibu-screenshot-1.png
# source url to kaibu.org
source: https://kaibu.org
# add custom badge
badge:
- icon: https://imjoy.io/static/badge/launch-imjoy-badge.svg
label: Launch ImJoy
url: https://imjoy.io/#/app?plugin=https://kaibu.org/#/app
For more application examples, see the manifest for ImJoy.
Describing datasets, notebooks and other types
The RDF allows for the description of datasets (type=dataset
), notebooks (type=notebook
) and other potential resources, you can use set source
and/or download_url
to point to the resource, or use attachments
to specify a list of associated files.
For examples, see entries dataset
/notebook
in the ZeroCostDL4Mic collection.
Describing models with the unspecific RDF(not recommended, use the Model RDF instead)
In general, it is discouraged to use the general RDF to describe AI models and we recommend to follow the model RDF spec instead. However, in some cases, it is not possible to provide detailed fields defined in the model RDF spec, the general RDF can be used for discribing AI models.
To do that, you need to first set the type
field to model
.
A basic integration would be simply provide a download_url
to a zip file (for example, with the model weights, source code or executable binary file) hosted on Github releases, Dropbox, Google Drive etc. For example:
download_url: https://zenodo.org/record/3446812/files/unet2d_weights.torch?download=1
If the model is available as a github repo, then provide the git_repo
field:
git_repo: https://github.com/my/model...
Here an example of a general RDF describing a model (not recommended): https://github.com/CellProfiling/HPA-model-zoo/blob/2f668d87defddc6c7cd156259a8be4146b665e72/manifest.bioimage.io.yaml#L33-L59
Model Resource Description File Specification
Besides the general RDF spec, the Model Resource Description File Specification
(model RDF
) defines a file format for representing pretrained AI models in YAML format. This format is used to describe models hosted on the BioImage.IO model repository site.
Here is a list of model RDF Examples:
Collection Resource Description File Specification
Another specialized RDF spec, the Collection Resource Description File Specification
(collection RDF
) defines a file format for representing collections of resources for the BioImage.IO website.
Link between resource items (under development)
You can use links
which is a list of id
to other resource items, for example, if you want to associate an applicaiton with a model, you can set the links field of the models like the following:
application:
- id: HPA-Classification
source: https://raw.githubusercontent.com/bioimage-io/tfjs-bioimage-io/master/apps/HPA-Classification.imjoy.html
model:
- id: HPAShuffleNetV2
source: https://raw.githubusercontent.com/bioimage-io/tfjs-bioimage-io/master/models/HPAShuffleNetV2/HPAShuffleNetV2.model.yaml
links:
- HPA-Classification
Hosting the file (under development)
It is recommended to host the resource description file on one of the public git repository website, including Github, Gitlab, Bitbucket, or Gist. A link can be submitted to BioImage.IO so we can keep track of the changes later.
Recommendations
- For AI models, consider using the model-specific spec (i.e. model RDF) instead of the general RDF. Only fallback to the general RDF if writing model specific RDF is not possible for some reason.
- The RDF or package file name should not contain spaces or special characters, it should be concise, descriptive, in kebab case or camel case.
- 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 file and try to avoid using subdirectories.
- Use the bioimage.io spec validator to verify your YAML file
- Store the yaml file in a version controlled Git repository (e.g. Github or Gitlab)
- Use or upgrade to the latest format version
BioImage.IO Spec Validator
It is recommended to use our validator to verify your models when you write it manually or develop tools for generating RDF/MDF files.
The spec validator can be installed either with pip
, or conda
:
# pip
pip install -U bioimageio.spec
# conda
conda install -c conda-forge bioimageio.spec
To use the spec validator, you can verify a model configuration in the bioimage.io model format using the following command:
bioimageio validate <MY-MODEL>.yaml
The output of this command will indicate missing or invalid fields in the model file. For example, if the field timestamp
was missing it would print the following:
{'timestamp': ['Missing data for required field.']}
or if the field test_inputs
does not contain a list, it would print:
{'test_inputs': ['Not a valid list.']}.
Changelog
bioimageio.spec 0.4.1.post5
- fixes to collection RDF:
- RDFs specified directly in collection RDF are validated correctly even if their source field does not point to an RDF.
- nesting of collection RDF allowed
bioimageio.spec 0.4.1.post4
- fixed missing field
icon
in general RDF's raw node - fixes to collection RDF:
- RDFs specified directly in collection RDF are validated correctly
- no nesting of collection RDF allowed for now
links
is no longer an explicit collection entry field ("moved" to unknown)
bioimageio.spec 0.4.1.post0
- new model spec 0.3.5 and 0.4.1
bioimageio.spec 0.4.0.post3
load_raw_resource_description
no longer acceptsupdate_to_current_format
kwarg (useupdate_to_format
instead)
bioimageio.spec 0.4.0.post2
load_raw_resource_description
acceptsupdate_to_format
kwarg
RDF Format Versions
collection RDF 0.2.1
- First official release, extends general RDF with fields
application
,model
,dataset
,notebook
and (nested)collection
, which hold lists linking to respective resources.
(general) RDF 0.2.1
- Non-breaking changes
- add optional
email
andgithub_user
fields to entries inauthors
- add optional
maintainers
field (entries like inauthors
butgithub_user
is required (andname
is not))
- add optional
model RDF 0.4.1
- Breaking changes that are fuly auto-convertible
- moved field
dependencies
toweights:pytorch_state_dict:dependencies
- moved field
- Non-breaking changes
documentation
field accepts URLs as well
model RDF 0.3.5
- Non-breaking changes
documentation
field accepts URLs as well
model RDF 0.4.0
- Breaking changes
- model inputs and outputs may not use duplicated names.
- model field
sha256
is required ifpytorch_state_dict
weights are defined. and is now moved to thepytroch_state_dict
entry asarchitecture_sha256
.
- Breaking changes that are fully auto-convertible
- model fields language and framework are removed.
- model field
source
is renamedarchitecture
and is moved together withkwargs
to thepytorch_state_dict
weights entry (if it exists, otherwise they are removed). - the weight format
pytorch_script
was renamed totorchscript
.
- Other changes
- model inputs (like outputs) may be defined by
scale
ing andoffset
ing areference_tensor
- a
maintainers
field was added to the model RDF. - the entries in the
authors
field may now additionally containemail
orgithub_user
. - the summary returned by the
validate
command now also contains a list of warnings. - an
update_format
command was added to aid with updating older RDFs by applying auto-conversion.
- model inputs (like outputs) may be defined by
model RDF 0.3.4
- Non-breaking changes
- Add optional parameter
eps
toscale_range
postprocessing.
- Add optional parameter
model RDF 0.3.3
- Breaking changes that are fully auto-convertible
reference_input
for implicit output tensor shape was renamed toreference_tensor
model RDF 0.3.2
- Breaking changes
- The RDF file name in a package should be
rdf.yaml
for all the RDF (notmodel.yaml
); - Change
authors
andpackaged_by
fields from List[str] to List[Author] with Author consisting of a dictionary{name: '<Full name>', affiliation: '<Affiliation>', orcid: 'optional orcid id'}
; - Add a mandatory
type
field to comply with the general RDF. Only valid value is 'model' for model RDF; - Only allow
license
identifier from the SPDX license list;
- The RDF file name in a package should be
- Other changes
- Add optional
version
field (default 0.1.0) to keep track of model changes; - Allow the values in the
attachments
list to be any values besides URI;
- Add optional
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for bioimageio.spec-0.4.1.post5.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 150f25e3dab7272cbf2121d9aea8fcda0ce5d44b63bd4cac925be2d10b6d98d4 |
|
MD5 | f40b21429f5ca3b0ede47713222775df |
|
BLAKE2b-256 | cee3a1fc3b3e6db2c3084f05bef5d3c47f69a7a56196390ef9def99f4750536b |
Hashes for bioimageio.spec-0.4.1.post5-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | a054fdc4177b61b6e18a86b11cfc71b7e32c69814dcd98eb72b8e783e784ca56 |
|
MD5 | 4eb24ef308affdea22e0b06f76a6b5b7 |
|
BLAKE2b-256 | e292e2fbe95f78fee776ffe1a2a56e3bf6b0fb2d06e1d1e5f3dc692f11e4a8d2 |