v0.3.3

What's Changed

• Imagenet loading improvements suggested by @sayakpaul by @ianstenbit in #774
• Add TPU support for basic_training by @ianstenbit in #757
• Add accelerator count to training history by @ianstenbit in #776
• Porting EfficientNets to KerasCV by @AdityaKane2001 in #740
• Promote model tests to integration by @LukeWood in #788
• Add pre-trained weights for DenseNet201 by @ianstenbit in #785
• Allow the KerasCV AnchorBox generator to work the image_shape instead of just images by @LukeWood in #790
• Add segmask support for RandomRotate by @ianstenbit in #780
• Add API to disable geometric transformations in RandAugment by @LukeWood in #792
• Some final touches to the RetinaNet API by @LukeWood in #784
• Step 1/2: refactor kerascv by @tanzhenyu in #791
• RandomFlip : added test for ragged bounding box by @divyashreepathihalli in #789
• Add segmask support for RandomResizedCrop by @ianstenbit in #777
• Add box matcher. by @tanzhenyu in #794
• Fixed samplers to support tensor shapes by @quantumalaviya in #798
• add reduction=none parameter to losses to fix retinanet detection example by @akucia in #796
• Update anchor_generator.py to fix docstring by @LukeWood in #800
• Fix bounding_box.clip_to_image() by @LukeWood in #802
• Add base feature pyramid network (FPN) to Keras CV by @qlzh727 in #803
• Add ROIGenerator. by @tanzhenyu in #804
• Add within 3d box detection. by @tanzhenyu in #787
• Add target gather as private API. by @tanzhenyu in #813
• Update docs for contrastive projector and probe by @ianstenbit in #814
• Add ROIPooler layer for RCNN. by @tanzhenyu in #808
• Setup initial benchmarks for model training by @ianstenbit in #810
• added the implementations of resnet 18 and 34 by @zaccharieramzi in #805
• Add classification head for segmentation model by @qlzh727 in #817
• add coordinate transform and spherical transform. by @tanzhenyu in #819
• Split RandomlyResizedCrop into two API surfaces (RandomlyZoomedCrop, RandomCropAndResize) by @AdityaKane2001 in #738
• Add ROISampler. by @tanzhenyu in #818
• Add Stable Diffusion by @fchollet in #828
• Add mixed precision support by @LukeWood in #834
• Stable diffusion small fixes by @bhack in #836
• Do image clipping in numpy space by @fchollet in #838

New Contributors

• @akucia made their first contribution in #796
• @zaccharieramzi made their first contribution in #805
• @fchollet made their first contribution in #828

Full Changelog: v0.3.2...v0.3.3

v0.3.2

What's Changed

• Add tensorflow_datasets dep to Dockerfile by @bhack in #750
• Add all of the no-op segmentation mask augmentations by @ianstenbit in #751
• Promote anchor generator default to a public api by @LukeWood in #755
• Write steps_per_second in basic_training by @ianstenbit in #758
• Remove cd to keras_Cv by @bhack in #756
• Fix FocalLoss reduction, SmoothL1 reduction, and add extra loss validation to RetinaNet model by @LukeWood in #760
• Fix loss reduction in the RetinaNet trainstep by @LukeWood in #764
• Add integration tests to GCB by @LukeWood in #761
• Update object_detection_base_model.py by @LukeWood in #771
• SimCLR implementation by @ianstenbit in #699
• Remove the tensorflow_datasets as installation dep by @qlzh727 in #772
• Remove the tfds (typo) from the requirements.txt by @qlzh727 in #773
• Fix weight loading in RetinaNet and introduce unit tests to ensure weight loading works as intended by @LukeWood in #768
• Add segmask support for random_flip by @ianstenbit in #775

Full Changelog: v0.3.1...v0.3.2

v0.3.1

What's Changed

• Improves shell/format.sh for local environment setups by @atuleu in #658
• Add v0 pre-trained weights for DenseNet169 by @ianstenbit in #660
• Restructure training scripts to avoid duplication by @ianstenbit in #659
• Introduce an Object Detection API to KerasCV by @LukeWood in #609
• Use aliasing for weights config to avoid duplicate weight files by @ianstenbit in #661
• Added no-op augment_bounding_boxes() methods by @divyashreepathihalli in #656
• Porting over RegNets from keras.applications by @AdityaKane2001 in #590
• Revert "Porting over RegNets from keras.applications (#590)" by @LukeWood in #674
• applications -> models by @LukeWood in #683
• Add IoU loss by @quantumalaviya in #634
• Make initial learning rate flag-controlled in basic_training by @ianstenbit in #687
• Add a contribution guide for training scripts by @ianstenbit in #626
• Fix classes docstring to match code behavior by @ianstenbit in #689
• Add GPU-accelerated testing by @ianstenbit in #692
• Fix flag type for initial_learning_rate by @ianstenbit in #694
• Add script version to training history script by @ianstenbit in #688
• Add initial ResNet50V2 weights by @ianstenbit in #695
• Switch to SGD for basic_training by @ianstenbit in #698
• Introduce re-usable ObjectDetection components to KerasCV by @LukeWood in #686
• Removed bias from Conv2D if followed by BatchNorm. by @sebastian-sz in #697
• Add MobileNetV3 models by @Rishit-dagli in #526
• Introduce a PascalVOC loader as an API by @LukeWood in #703
• skip tests to try to prevent OOMs by @LukeWood in #708
• Add TFDS+Pandas as test deps for GPU tests by @ianstenbit in #709
• Add augmenter layer by @ianstenbit in #704
• Move loaders namespace to datasets for API consistency by @LukeWood in #711
• Add instructions for adding test deps by @ianstenbit in #714
• Encode labels in the tf.data pipeline for Object Detection models by @LukeWood in #705
• Add ragged bounding box support and utils to pad and filter ragged bounding boxes to keras cv by @divyashreepathihalli in #677
• Make ReduceLrOnPlateau a bit less patient for basic imagenet training by @ianstenbit in #723
• update demo scripts by @divyashreepathihalli in #722
• Update the API & train_step() of RetinaNet to encourage TPU compatible forward passes by @LukeWood in #721
• Add ConvNeXt by @sayakpaul in #678
• Restructure model tests to try to reduce OOM by @ianstenbit in #736
• Create ImageNet loader in keras_cv.datasets by @ianstenbit in #735
• Add improved ResNet50V2 weights by @ianstenbit in #741
• Support validation_data in fit() for KerasCV RetinaNet by @LukeWood in #743
• Add support for model kwargs to basic training script by @ianstenbit in #742
• Make loss reduction consistent with model garden by @LukeWood in #747
• Add support for segmentation mask in BaseImageAugmentationLayer by @ianstenbit in #748

New Contributors

• @Rishit-dagli made their first contribution in #526

Full Changelog: v0.2.10...v0.3.1

v0.2.10

What's Changed

• Add a test case to enforce rescaling behavior by @ianjjohnson in #603
• Improves serialization_test.py to cover all argument with default values by @atuleu in #600
• Porting over ResNets from keras.applications by @AdityaKane2001 in #558
• Robust serialization test by @AdityaKane2001 in #591
• Add clip_to_image() util to bounding_box by @divyashreepathihalli in #607
• fix random_rotation bugs by @divyashreepathihalli in #608
• classifier_activation docstring corrections by @AdityaKane2001 in #610
• Fixed converters so that rel->rel don't require images by @quantumalaviya in #612
• Randomshear Bounding box augmentation Feature by @pranavjadhav001 in #373
• Style changes to NMS layer by @LukeWood in #613
• Added random_rotation_demo.py by @divyashreepathihalli in #555
• Add a link to the new COCO metric publication & keras.io docs by @LukeWood in #620
• Migrate iou_utils to the keras_cv.bounding_box namespace by @quantumalaviya in #598
• code style fixes in random_rotation.py by @divyashreepathihalli in #624
• Fixed error message when requirement does not meet for minimum TF version by @adhadse in #563
• Add an initial Densenet/Imagenet classification training script by @ianjjohnson in #618
• Fix NMSLayer shape inference to work when no elements are present by @LukeWood in #631
• Fixed reference sections in some docstrings by @quantumalaviya in #633
• Add pre-trained weight loading (starting with DenseNet121) by @ianjjohnson in #627
• Update basic_training example to only save the best weights by @ianjjohnson in #635
• Add initial weight exporting scripts by @ianjjohnson in #628
• Adding new NMS tests by @quantumalaviya in #642
• Add input-tensor param to all KerasCV.models by @ianjjohnson in #639
• Use fstring format in models/weights.py by @ianjjohnson in #645
• Silence E402 linter error by @ianjjohnson in #651
• added RandomFlip to Keras_CV by @divyashreepathihalli in #562
• Switch to tensorboard.dev for training metadata handling by @ianjjohnson in #648
• Add Focal Loss by @quantumalaviya in #615
• Cleanup and polish on basic_training.py by @ianjjohnson in #652
• Rename num_classes to classes across entire repo by @ianjjohnson in #657

New Contributors

• @adhadse made their first contribution in #563

Full Changelog: v0.2.9...v0.2.10

v0.2.9

What's Changed

• Added parameterized tests to RRC by @AdityaKane2001 in #550
• Add seeds for random distribution sampling by @quantumalaviya in #557
• Add support for unbatch bounding boxes to convert_format() by @LukeWood in #556
• Squeeze and Excite block by @AdityaKane2001 in #505
• Fixed #425 with Lambda layer by @AdityaKane2001 in #559
• Add default name to DenseNet (to match docstring) by @ianjjohnson in #574
• [Devcontainer] Fix isort extension name by @bhack in #569
• Added visualization feature to demo utils by @divyashreepathihalli in #567
• Adding Non-Max Suppression Layer by @quantumalaviya in #564
• Migrate VGG19 to keras_cv.models by @ianjjohnson in #573
• Adds support for ragged tensor to bounding_box.convert_format by @atuleu in #572
• Update NMS Layer file location by @LukeWood in #576
• Minor edits to make bounding box utilities more idiomatic by @LukeWood in #577
• Add serialization test for NMS by @quantumalaviya in #578
• Changed BaseImageAugmentationLayer to preserve additional unmodified … by @sebastian-sz in #568
• Add MLP Mixer models by @sayakpaul in #529
• Towards keypoint support: BaseImageAugmentationLayer modification by @atuleu in #586
• Add CSPDarkNet, DarkNet21 & DarkNet53 by @quantumalaviya in #541
• Towards keypoint support: Converter and utility functions by @atuleu in #585
• Include with_targets to RandomResizedCrop by @LukeWood in #597

New Contributors

• @atuleu made their first contribution in #572

Full Changelog: v0.2.8...v0.2.9

v0.2.8

What's Changed

• Roadmap by @LukeWood in #498
• Stochastic Depth argument name correction by @AdityaKane2001 in #497
• Correction in layers.__init__.py by @AdityaKane2001 in #503
• Migrate RandomRotation to Keras_CV by @divyashreepathihalli in #493
• random jpeg quality by @artu1999 in #437
• added yxyx and rel_yxyx bounding box format support in bounding box utils by @divyashreepathihalli in #510
• Remove "tf.keras.internal.layers" from all files by @AdityaKane2001 in #516
• Drop Path by @AdityaKane2001 in #480
• add: PR description template. by @sayakpaul in #525
• Add an augment_label method to the RandomGaussianBlur augmentation layer by @ianjjohnson in #537
• Update all preprocessing layer demos to use demo_utils by @ianjjohnson in #535
• Remove references to data_format by @ricardoprins in #531
• Fix DenseNet Docstring by @quantumalaviya in #539
• Change all call arguments using tf.float32 as a default to the string "float32" by @ricardoprins in #542
• Update serialization_test.py by @AdityaKane2001 in #544
• Random Resized Crop by @AdityaKane2001 in #499

New Contributors

• @ianjjohnson made their first contribution in #537
• @ricardoprins made their first contribution in #531

Full Changelog: v0.2.7...v0.2.8

v0.2.7

What's Changed

• Stochastic depth by @AdityaKane2001 in #474
• chore: minor updates to doc. by @sayakpaul in #481
• fix typo in Devcontainer by @bhack in #483
• ref(examples): Refactored example helper functions by @kartik4949 in #405
• Migrate BaseImageAugmentationLayer to Keras_CV by @divyashreepathihalli in #482
• Add AugMix by @quantumalaviya in #407
• Fix major numerical issue in COCOMeanAveragePrecision by @LukeWood in #492
• Creates a prototype of a model for keras_cv.models. Used DenseNet as an example by @LukeWood in #476

New Contributors

• @AdityaKane2001 made their first contribution in #474
• @sayakpaul made their first contribution in #481
• @divyashreepathihalli made their first contribution in #482

Full Changelog: v0.2.6...v0.2.7

v0.2.6 Release

What's Changed

• cut_mix example: tf.cast label from int to float by @johnypark in #445
• Implement clear error messages to CutMix and MixUp by @LukeWood in #453
• Update the CI to TF stable by @bhack in #463
• Add pre-commit hooks by @bhack in #462
• Fixing Equalization Shape Calculation During Trace Time by @quantumalaviya in #475
• Implement the new bounding box API by @LukeWood in #441

New Contributors

• @johnypark made their first contribution in #445

Full Changelog: v0.2.5...v0.2.6

v0.2.5

What's Changed

• Introduces MaybeApply layer. by @sebastian-sz in #435
• Add get_config() to COCOMetrics and add test coverage by @LukeWood in #444

Full Changelog: v0.2.4...v0.2.5

v0.2.4

update bounding box entrypoint from utils.bounding_box -> bounding_box