1.1.0-alpha.1

What's Changed

• Add random and foreground oversampling in ROIDataset by @perctrix in #83
• SegmentationTrainer visualization suppresses small values by @ATATC in #121
• soft_dice_coefficient incorrectly flattens all classes together by @perctrix in #125

Full Changelog: 1.1.0-alpha.0...1.1.0-alpha.1

1.1.0-alpha.0

What's Changed

• Support loading a checkpoint in Trainer by @ATATC in #108
• Multiple values for self in Trainer.save_everything_for_recovery() by @ATATC in #112
• Incorrect unpacking order of paddings in Restore3d causes dimension mismatch by @perctrix in #115
• TrainerTracker.worst_case involves torch.Tensor that causes asdict() failure by @ATATC in #116
• Support optionally saving and loading toolbox by @ATATC in #117

Full Changelog: 1.0.2...1.1.0-alpha.0

1.0.2

What's Changed

• Make paths() and save_paths() a feature of all path-based datasets by @ATATC in #75
• Support training on variable-sized medical images with random patch sampling by @ATATC in #76
• Unused imports in "metrics.py" by @ATATC in #85
• Add ROI shape validation by @perctrix in #79
• Add a helper that determines the optimal device by @ATATC in #88
• Use auto_device() in sanity_check() when device is not specified by @ATATC in #90
• Add __str__() for Trainer by @ATATC in #92
• Add SimpleDataset by @ATATC in #94
• Support extracting filenames in predictors from PathBasedUnsupervisedDataset by @ATATC in #96
• Inference failed saving prediction by @ATATC in #97
• Incorrect dimension in statistical_foreground_shape() for 3D data by @ATATC in #101
• LayerT.update() should return a new instance by @ATATC in #103
• Expose APIs in Trainer by @ATATC in #100
• load_inspection_annotations() is completely non-functional by @ATATC in #86
• Add ComposeDataset that takes multiple datasets and merges them into one by @ATATC in #105
• DiceBCELossWithLogits is expected to support multiclass by @ATATC in #110

Full Changelog: 1.0.1...1.0.2

1.0.2-beta.1

What's Changed

• Incorrect dimension in statistical_foreground_shape() for 3D data by @ATATC in #101
• LayerT.update() should return a new instance by @ATATC in #103
• Expose APIs in Trainer by @ATATC in #100
• load_inspection_annotations() is completely non-functional by @ATATC in #86
• Add ComposeDataset that takes multiple datasets and merges them into one by @ATATC in #105

Full Changelog: 1.0.2-beta.0...1.0.2-beta.1

1.0.2-beta.0

What's Changed

• Unused imports in "metrics.py" by @ATATC in #85
• Add ROI shape validation by @perctrix in #79
• Add a helper that determines the optimal device by @ATATC in #88
• Use auto_device() in sanity_check() when device is not specified by @ATATC in #90
• Add __str__() for Trainer by @ATATC in #92
• Add SimpleDataset by @ATATC in #94
• Support extracting filenames in predictors from PathBasedUnsupervisedDataset by @ATATC in #96
• Inference failed saving prediction by @ATATC in #97

Full Changelog: 1.0.2-alpha.0...1.0.2-beta.0

1.0.2-alpha.0

What's Changed

• Make paths() and save_paths() a feature of all path-based datasets by @ATATC in #75
• Support training on variable-sized medical images with random patch sampling by @ATATC in #76

Full Changelog: 1.0.1...1.0.2-alpha.0

1.0.1

What's Changed

• Support editing frontend secrets through a command line tool by @ATATC in #2
• Rename mipcandy.preset to mipcandy.presets by @ATATC in #6
• Modularize the getter of model complexity info by @ATATC in #11
• Enhance sanity_check() by @ATATC in #13
• Add K-fold cross-validation support by @ATATC in #18
• Support loading default values of arguments of Trainer.train() from settings by @ATATC in #22
• Log errors in Trainer by @ATATC in #24
• 3D segmentation training blocked by PyVista rendering windows after each epoch by @ATATC in #19
• Add multimodal support for NNUNetDataset by @ATATC in #20
• Multi-channel 3D visualization crashes in SegmentationTrainer._save_preview() by @ATATC in #29
• SegmentationTrainer crashes because "min_all" not implemented for 'UInt16' during preview generation by @perctrix in #31
• Fix SlidingWindow 3D metadata dimensions causing revert failure by @perctrix in #35
• Fix preserve subclass type in NNUNetDataset.fold() method by @perctrix in #39
• Add SlidingSegmentationTrainer for consistent shapes by @ATATC in #42
• Set the model to inference mode during sanity check by @ATATC in #46
• Create a type alias for torch.device | str by @ATATC in #51
• Use quotient regression for vlidation score prediction instead of polynomial regression by @ATATC in #48
• Support estimated time of completion prediction by @ATATC in #56
• Support updating keyword arguments in LayerT by @ATATC in #58
• Make the console prettier by using rich more by @ATATC in #60
• TypeError: 'NoneType' object is not subscriptable in predict_maximum_validation_score() when validation score prediction fails by @ATATC in #63
• Frontend metrics display raw values without unit conversion, causing unreadable numbers by @ATATC in #61
• Redundant logging in train() by @ATATC in #65
• Rename "mask" to "output" for generic usage by @ATATC in #67
• visualize3d() makes training freeze when saving 3D previews due to incorrect PyVista off-screen timing by @perctrix in #55
• Include date in ETC by @ATATC in #69
• Support built-in dataset downloading by @ATATC in #70
• Add multi-output support in sanity_check() by @ATATC in #72

New Contributors

• @ATATC made their first contribution in #2

Full Changelog: 1.0.0...1.0.1

1.0.1-beta.2

What's Changed

• Make the console prettier by using rich more by @ATATC in #60
• TypeError: 'NoneType' object is not subscriptable in predict_maximum_validation_score() when validation score prediction fails by @ATATC in #63
• Frontend metrics display raw values without unit conversion, causing unreadable numbers by @ATATC in #61
• Redundant logging in train() by @ATATC in #65
• Rename "mask" to "output" for generic usage by @ATATC in #67
• visualize3d() makes training freeze when saving 3D previews due to incorrect PyVista off-screen timing by @perctrix in #55
• Include date in ETC by @ATATC in #69
• Support built-in dataset downloading by @ATATC in #70

Full Changelog: 1.0.1-beta.1...1.0.1-beta.2

1.0.1-beta.1

What's Changed

• Set the model to inference mode during sanity check by @ATATC in #46
• Create a type alias for torch.device | str by @ATATC in #51
• Use quotient regression for vlidation score prediction instead of polynomial regression by @ATATC in #48
• Support estimated time of completion prediction by @ATATC in #56
• Support updating keyword arguments in LayerT by @ATATC in #58

Full Changelog: 1.0.1-beta.0...1.0.1-beta.1

1.0.1-beta.0

What's Changed

• Modularize the getter of model complexity info by @ATATC in #11
• Enhance sanity_check() by @ATATC in #13
• Add K-fold cross-validation support by @ATATC in #18
• Support loading default values of arguments of Trainer.train() from settings by @ATATC in #22
• Log errors in Trainer by @ATATC in #24
• 3D segmentation training blocked by PyVista rendering windows after each epoch by @ATATC in #19
• Add multimodal support for NNUNetDataset by @ATATC in #20
• Multi-channel 3D visualization crashes in SegmentationTrainer._save_preview() by @ATATC in #29
• SegmentationTrainer crashes because "min_all" not implemented for 'UInt16' during preview generation by @perctrix in #31
• Fix SlidingWindow 3D metadata dimensions causing revert failure by @perctrix in #35
• Fix preserve subclass type in NNUNetDataset.fold() method by @perctrix in #39
• Add SlidingSegmentationTrainer for consistent shapes by @ATATC in #42

Full Changelog: 1.0.1-alpha.0...1.0.1-beta.0