0.12.0

🚀 Features

Update for new tracking approach @MekWarrior (#168)

What

• Updates for DeepCell's approach to the tracking problem in live-cell imaging. The Tracking Application has been updated and now requires two different models.

Why

• These updates represent the natural evolution of DeepCell's approach to tracking. It addresses key needs to further generalize the codebase and dramatically improves tracking speed.

Add new `detect_dimension_order` to detect the dim ordering string. @willgraf (#172)

Detects the dimension ordering for 2D and 3D data for channels first or channels last, and placed in Redis under the key dim_order.

This facilitates handing off the dim ordering to downstream visualization services.

Fixes #169

🐛 Bug Fixes

Define batch size and pass to app.predict. @willgraf (#175)

app.predict no longer supports batch_size=None.

Ignore negative batch dimensions when calculating batch\_size. @willgraf (#173)

Fixes #167

Ignore all negative dimensions when calculating batch size for the GrpcModelWrapper. This PR ensures that the batch size calculated will never be negative, but does not ensure that the batch size calculated will be optimal if there are many unknown dimensions.

🧰 Maintenance

Create GitHub Workflow for the release-drafter. @willgraf (#171)

Add release-drafter template file. @willgraf (#170)

0.11.0

0.11.0

Features

• Allow optional channels argument to dictate which channel axes to use for inference. (#162)
• Add new MAX_IMAGE_FRAMES to prevent the TrackingConsumer from running out of memory for large inputs. (#166)

Bugfixes

• Validate images are smaller than MAX_IMAGE_HEIGHT and MAX_IMAGE_WIDTH (#163)
• Add prediction status update to MesmerConsumer. (#164)

0.10.0

0.10.0

Features

• Update to deepcell 0.9.0 and migrate MultiplexConsumer to MesmerConsumer (#160).

Breaking Changes

• Drop support for Python 3.5.

0.9.0

0.9.0

Features

• Leverage deepcell.applications to package deep learning workflows using a GrpcModelWrapper. (#154)

Breaking Changes

• Drop support for Python2.7 and introduce deepcell-cpu as a dependency.
• ImageFileConsumer has been refactored to SegmentationConsumer.
• GKE_STORAGE_BUCKET and AWS_S3_BUCKET have been replaced with STORAGE_BUCKET which requires a bucket protocol (s3://bucketname). Remove CLOUD_PROVIDER.
• DEBUG has been replaced with LOG_LEVEL.
• Replace custom protos with tensorflow-serving-api and tensorflow APIs.

0.8.4

0.8.4

Features

• Split PyTest and Coveralls into separate build stages, and skip coveralls if the token is not set. (#157)

Bugfixes

• Update TrackingConsumer for TensorFlow 2 based model (#156)
• Revert output filename pattern (#153)

0.8.3

0.8.3

Features

• Mock MultiplexConsumer postprocessing during tests to reduce 80% of test time. (#151)

Bugfixes

• Prevent OOM errors by raising an error if the scale provided is outside MAX_SCALE and MIN_SCALE (#147)
• Prevent rescaled images from being returned as a different size. (#149)

0.8.2

0.8.2

Features

• Allow environment variable definitions for pre-processing, similar to post-processing. (#146)

Bugfixes

• Bump deepcell_toolbox to 0.8.2.

0.8.1

0.8.1

Bugfixes

• Fixed docker deploymet GitHub Action to run for all published releases, not just created ones. (#143)
• Consumers return all data in the output signature, no longer looking for the name "prediction". (#144)

0.8.0

0.8.0

Features

• Only process images smaller than MAX_IMAGE_HEIGHT and MAX_IMAGE_WIDTH. (#128)
• Improve Multiplex image validation and reduce required heads for post-processing (#125, #130, #131, #132)
• Migrate from Travis CI to GitHub Actions (#140, #141)

Bugfixes

• Update deepcell-toolbox and deepcell-tracking to latest patches (#126)

0.7.0

0.7.0

Features

• Added MultiplexConsumer to support multichannel cell segmentation (#121)
• Improved testing with pytest-mock (#117)
• Improved documentation for custom consumers (#120)
• Updated copyright year (#118, #122)