v1.7.0 🌈

This release contains several improvements to automatic instance segmentation and other features. As part of these improvements micro_sam.instance_segmentation.get_amg was renamed to micro_sam.instance_segmentation.get_instance_segmentation_generator.

Changes

• Release 1.7 @constantinpape (#1134)
• Add support for training from in-memory data @constantinpape (#1135)
• Automatic prompt generator @constantinpape (#1122)
• Update image preprocessing @constantinpape (#1133)
• Masked segmentation @constantinpape (#1131)
• Refactor APG interface and instance segmentation tests @constantinpape (#1127)
• Remove additional batch size checks in training @anwai98 (#1129)
• Fix training/validation metrics scaling and validation loss aggregation @YonghaoZhao722 (#1126)
• Add support for loading initial segmentation results to series annotator @constantinpape (#1125)
• Use stability score in nms @constantinpape (#1123)
• add sam_model as attribute for torch.nn.Module magic to work @FynnBe (#1111)
• Minor fix to 3d encoder wrapper for expected model type configuration @anwai98 (#1110)
• Update RELEASE_OVERVIEW.md @anwai98 (#1109)

v1.6.2 🌈

Changes

• New release @constantinpape (#1108)
• Fix image series annotator @constantinpape (#1107)
• Add model versions and create cache directory for info CLI @anwai98 (#1106)
• Release new medical imaging models @anwai98 (#1103)
• Update RELEASE_OVERVIEW.md @anwai98 (#1098)

v1.6.1 🌈

Changes

• Bump version to 1.6.1 @constantinpape (#1097)
• Merge dev to master @anwai98 (#1091)
• Minor fix to additional arguments for training instance segmentation @anwai98 (#1096)
• Add is_multi_tensor argument to default_sam_dataset to split kwargs in loader @lufre1 (#1092)
• Update default_sam_dataset to allow parsing expected parameters @lufre1 (#1090)
• Add support scripts for training custom model @anwai98 (#1086)
• Add script for counting total images and annotated objects @anwai98 (#1082)
• Add trackastra as dependency @constantinpape (#1081)
• Update grid search functionality for more flexibility @constantinpape (#1079)
• Add new BAND site @anwai98 (#1080)
• Minor linting fixes @anwai98 (#1078)
• More doc updates @anwai98 (#1068)
• Update RELEASE_OVERVIEW.md @anwai98 (#1067)

v1.6.0 🌈

With this version we publish improved versions of the light microscopy and electron microscopy models.

Changes

• Bump version @constantinpape (#1066)
• Update embedding visualization and add support for ROIs in SAM datasets @constantinpape (#1065)
• Prepare new release @constantinpape (#1064)
• Allow support for custom label transforms in default_sam_dataset @lufre1 (#1063)
• Start preparing release for new models @anwai98 (#1061)
• Update documentation for BAND @anwai98 (#1060)
• Minor updates to EM evaluation benchmarking scripts @anwai98 (#1059)
• More zarr v3 fixes @constantinpape (#1057)
• Update RELEASE_OVERVIEW.md @anwai98 (#1055)
• Update generalist training scripts @anwai98 (#1052)
• Update environment.yaml @anwai98 (#1050)

v1.5.0 🌈

Changes

• New release @constantinpape (#1049)
• Try to re-activate linux gui tests @constantinpape (#914)
• Update automatic tracking for storing tracks in CTC format @anwai98 (#1045)
• Resume tests for py3.12 @anwai98 (#1026)
• Fix issue in object prediction @constantinpape (#1048)
• First working prototype for object classifier @constantinpape (#1036)
• Add configuration for training on smaller GPUs @anwai98 (#1046)
• Change the QLoRA model export to work with Late QLoRA @caroteu (#1041)
• Minor update to instance segmentation finetuning - expose loss function @anwai98 (#1042)
• Remove context manager for zarr.open @constantinpape (#1043)
• Minor update to automatic segmentation CLI - verbosity argument @anwai98 (#1040)
• Make passing checkpoint flexible in livecell inference script @anwai98 (#1034)
• Ensure device consistency in onnx model export @anwai98 (#1029)
• Remove napari pin to allow v6 supports @anwai98 (#1019)
• Minor update to return value for iterative prompting evaluation @anwai98 (#1024)
• Add support for different prompt choice in interactive seg @anwai98 (#1022)
• Minor update to all docstrings @anwai98 (#1023)
• Minor updates to quick start notebook @anwai98 (#1016)
• Remove verbosity argument from learning rate schedulers @anwai98 (#1021)
• Add quick-start notebook @haesleinhuepf (#1013)
• Better handling of input prompts in first iteration @anwai98 (#991)
• Minor typo fix in documentation @anwai98 (#1010)
• Ensure valid layers exist for each annotator click menu @anwai98 (#943)
• Implement instance segmentation training and update training tests @constantinpape (#1008)
• Updates to automatic segmentation @constantinpape (#979)
• Add micromamba installation instructions in FAQ section @anwai98 (#1005)
• Update notebooks for updated conda installation @anwai98 (#1004)
• Update README.md @constantinpape (#1003)
• Minor fix to remove peft kwargs from evaluation scripts @anwai98 (#1002)
• Add zarr to yaml file @anwai98 (#996)
• Fix model export issues @anwai98 (#989)
• Update example scripts for automatic segmentation @anwai98 (#988)
• Add documentation for info and evaluation CLIs @anwai98 (#981)
• Minor update to avoid using napari pbar in CLI @anwai98 (#987)
• Minor fix to training CLI argument @anwai98 (#984)
• Update the documentation for v1.4.0 @constantinpape (#975)
• Update RELEASE_OVERVIEW.md @anwai98 (#974)

v1.4.0 🌈

Fix windows installation, GUI updates and automatic tracking

This release includes three main changes:

• Preliminary support for automatic tracking via Trackastra integration.
• Changes in the GUI to make the model names more informative
• Much easier installation on windows

Changes

• New release @constantinpape (#971)
• Minor update to environment file and other docs @anwai98 (#972)
• Fix automatic tracking with dask @constantinpape (#970)
• Update model combination tooltips and adapt other GUIs @anwai98 (#964)
• Better preservation for committed objects @constantinpape (#967)
• Update benchmark.py @YonghaoZhao722 (#966)
• Update commit to file @constantinpape (#959)
• Validate model type with model state instead of state @caroteu (#962)
• Minor update to training CLI @anwai98 (#957)
• Add scripts for annotating data on OMERO server @anwai98 (#941)
• Fix dependencies @constantinpape (#955)
• Add checks for valid values parsed to finetuning CLI arguments @anwai98 (#944)
• Omero support for TiM workshop @constantinpape (#946)
• Merge master into dev @constantinpape (#947)
• Start preparation for TIM2025 workshop @anwai98 (#925)
• Minor updates to embedding widget @anwai98 (#938)
• Tracking layer @constantinpape (#913)
• Fix issues in the image series annotator and load existing segmentati… @constantinpape (#936)
• Add download progress bar to napari info log @anwai98 (#933)
• Make GUI default model choice depend on micro_sam defaults @anwai98 (#934)
• Add validation window for avoiding recomputing embeddings @anwai98 (#790)
• Allow flexible loading of mismatching checkpoint-model type combination @anwai98 (#886)
• Add implementation for Late PEFT @caroteu (#916)
• Save outputs of automatic segmentation CLI in stages @anwai98 (#928)
• Implement batched prediction for the AIS segmentation decoder @constantinpape (#932)
• Update model choice design in annotator tools @anwai98 (#927)
• Add evaluation CLI mentions to usage FAQ @anwai98 (#930)
• Update GUI to avoid replicating existing label / prompt layers @anwai98 (#929)
• Implement batched embedding computation @constantinpape (#926)
• End: ensure the return of _to_image is a np.array (allows dask layers to work) @psobolewskiPhD (#922)
• Add support for avoiding to overwrite trained model @anwai98 (#919)
• Add information CLI @anwai98 (#911)
• Update environment file for installation @anwai98 (#912)
• Update embedding widget to pop dialog box for no images @anwai98 (#910)
• Support multiple images in automatic segmentation CLI @anwai98 (#904)
• Start implementation for automated tracking @constantinpape (#678)
• Add evaluation CLI for segmentations @anwai98 (#899)
• Update training section in doc with finetuning CLI mention @anwai98 (#900)
• Update example notebooks with comments on tiling-based inference @anwai98 (#901)
• Update default prompt propagation mode for multidimensional data @anwai98 (#903)
• Move automatic segmentation mode @constantinpape (#898)
• Remove exclusive yaml file for windows @anwai98 (#895)
• Update win installation WIP @constantinpape (#869)

v1.3.1 🌈

Changes

• Bump version @constantinpape (#885)
• Add support for preprocessing inputs in training CLI @anwai98 (#879)
• Update documentation with latest features @anwai98 (#890)
• Minor fix to handle is_seg_dataset argument @anwai98 (#892)
• Minor fix to automatic segmentation in 3d @anwai98 (#882)
• Add data submission section to the documentation @constantinpape (#881)
• Update RELEASE_OVERVIEW.md @anwai98 (#878)

v1.3.0 🌈

New Light Microscopy Models

This release introduces a new light microscopy model that was trained on a larger dataset and clearly improves automatic segmentation.

See the improvements (qualitative and quantitative) of the new model:

The vit-b version of this model is the new default model used in micro_sam.

Changes

• Bump version @constantinpape (#877)
• Add link to new models on BioImage.IO @anwai98 (#873)
• Minor update to remove debugging scripts @anwai98 (#876)
• Change default model @constantinpape (#874)
• Merge dev into master @constantinpape (#872)
• Release preparation for new LM v3 models @anwai98 (#862)
• Add menu option for automatic segmentation modes @anwai98 (#793)
• Update README.md @constantinpape (#871)
• Create RELEASE_OVERVIEW.md @anwai98 (#867)
• Update LM generalist scripts @anwai98 (#822)
• Refactor model export for training scripts @constantinpape (#864)
• Minor update to automatic segmentation mode defaults in CLI @anwai98 (#863)
• Add CLI for training SAM @anwai98 (#860)
• Add garbage collection after commit and clear annotations @anwai98 (#789)
• Extend support for GUI training on RGB images @anwai98 (#859)
• Extend automatic segmentation CLI to allow continuing annotation @anwai98 (#858)
• Minor updates to benchmarking CLI with new LM datasets @anwai98 (#861)
• Add scripts for promptable segmentation using weak labels @anwai98 (#857)

v1.2.2 🌈

Changes

• Update __version__.py @constantinpape (#856)
• Extend support for automatic seg functionality to return embeddings @anwai98 (#855)
• Minor updates to example script docs @anwai98 (#854)
• Allow flexible post-processing for larger segmentations @anwai98 (#853)

v1.2.1 🌈

This release implements several changes that are part of three of our recent publications that are built on top of micro_sam:

• medico-sam, which improves SAM for medical images
• peft-sam, which investigates parameter efficient finetuning for SAM
• patho-sam, which improves SAM for histopathology

Changes

• Prepare release 1.2.1 @constantinpape (#852)
• Parallel AIS implementation @constantinpape (#851)
• Disable GUI tests on Linux @constantinpape (#849)
• Add histopathology whole-slide image sample image @anwai98 (#848)
• Minor update to make rank optional in peft class @anwai98 (#846)
• Add support for tiling window in evaluation scripts @anwai98 (#844)
• Minor update to support peft kwargs in qualitative comparsion scripts @anwai98 (#845)
• Minor updates to UNETR model function @anwai98 (#843)
• Fix QLoRA weights and bias initialisation @caroteu (#833)
• Add medical imaging models to model registry @anwai98 (#841)
• Update dice weight constraints for flexible loss weighting @anwai98 (#839)
• Expose out_channels arguments for get_unetr function @anwai98 (#838)
• Add support for histopathology models @anwai98 (#834)
• Simplify logic for metric choice @anwai98 (#831)
• Extend dice score metric for multi-dim segmentation @anwai98 (#829)
• Add support for larger SAM models for 3d semantic segmentation @anwai98 (#827)
• Add QLoRA @anwai98 (#820)
• Minor update to prompt-based segmentation @anwai98 (#825)
• Minor fix to release overview doc @anwai98 (#821)
• Add release notes to doc @anwai98 (#819)