v0.8.0

This release integrates KerasCV with Kaggle Models. KerasCV models will now work in Kaggle offline notebooks and all assets will quickly attach to a notebook rather than needing a slow download.

Summary

Pre-trained KerasCV models are now available entirely through Kaggle Models, with the full list of models available in both KerasCV and KerasNLP hosted here.

Here is an example model page: EfficientNetV2
Note that each preset for the model is available via the Model Variations tab drop-down menu along with example usage.
Additionally, you can view the file structure containing the model architecture, metadata, and weights, if applicable.

This change will not affect the existing usage of from_preset(). Statements like keras_nlp.models.MobileNetV3Backbone.from_preset("mobilenet_v3_small") will continue to work and download checkpoints from the Kaggle Models hub.

An note on model saving—for saving support across Keras 2 and Keras 3, we recommend using the new Keras saved model format. You can use model.save('path/to/location.keras') for a full model and model.save_weights('path/to/location.weights.h5') for checkpoints. See the Keras saving guide for more details.

What's Changed

• Update the comment of the num_classes parameter of deeplab v3 by @aaudevart in #2071
• Set read-only tokens for all GitHub workflows by @pnacht in #2075
• Add the Segment Anything Model to KerasCV by @tirthasheshpatel in #1987
• Export SD submodels by @jbischof in #2080
• adding deeplab_v3_plus_presets by @divyashreepathihalli in #2051
• [Yolov8 detector] Fix example in docstring by @cosmo3769 in #2082
• Bump python version requirement to 3.9 by @mattdangerw in #2079
• [SAM] Fixed shape issue in example in docstring by @cosmo3769 in #2083
• Use "masks" and "boxes" as prompt keys for SAM by @ianstenbit in #2084
• Support partial prompting in SAM by @ianstenbit in #2085
• Explicitly specify full package for more modules, as shown in #2001 by @BjarneHerland in #2077
• Update VITDet to conform to KerasCV scaling standards by @ianstenbit in #2086
• Resolve mean/std swap for VITDet backbone by @ianstenbit in #2087
• Support importing Keras3 as Keras instead of Keras Core by @ianstenbit in #2089
• remove tests from pypi build by @haifeng-jin in #2094
• Export task models as both models.task.Model and models.Model by @ianstenbit in #2093
• Fix CSPDarkNetBackbone export by @ianstenbit in #2096
• Fix PyTorch tests by @ianstenbit in #2097
• Doc updates to switch branding to Keras 3 by @ianstenbit in #2092
• Fix a flaky test for Segment Anything by @ianstenbit in #2090
• Fix decoding of LabelEncoder for RetinaNet by @ianstenbit in #2099
• Use the proper title for example by @Philmod in #2109
• Fixes Fine-tuning Stable Diffusion by @ID6109 in #2113
• Fixed typo in README.md by @AniketP04 in #2106
• Update mask_invalid_detections.py by @VBPMP in #2117
• Fix the issue (#2102): Add cast to float32 from float16 into Stable D… in #2124
• Fix RetinaNet shape issue by @tirthasheshpatel in #2119
• update reviewers names by @divyashreepathihalli in #2130
• update keras-cv json file name by @divyashreepathihalli in #2128
• Improve keras 3 detection by @divyashreepathihalli in #2132
• Fix Segment Anything Model saving bug by @tirthasheshpatel in #2138
• Adds Nightly package for keras-cv by @sampathweb in #2139
• Set a security policy by @pnacht in #2142
• add Random ops shim by @divyashreepathihalli in #2145
• Replace RandomGenerator with SeedGenerator by @sampathweb in #2150
• Revert "Replace RandomGenerator with SeedGenerator" by @sampathweb in #2161
• Keras 3 compatibility by @divyashreepathihalli in #2170
• fix noise scheduler error in stable diffusion by @divyashreepathihalli in #2171
• Fix the issue(#2158): Add some file names into .gitignore by @y-vectorfield in #2159
• fix shapes error if images was a list. by @divyashreepathihalli in #2173
• Removed setup.py develop from documentation, replaced with pip instal… by @EdIzaguirre in #2118
• Remove keras-nightly pin by @sampathweb in #2174
• fix model None error by @divyashreepathihalli in #2176
• update version error by @divyashreepathihalli in #2175
• Fixes Saving Format related test failures for Keras 3 by @sampathweb in #2180
• convert to tensor before smart resize by @sampathweb in #2184
• Fix Keras 3 version check by @sampathweb in #2191
• Update version 0.8 by @sampathweb in #2190
• Fix bug when upranking passthrough inputs to RandAugment by @mattdangerw in #2194
• fix stable diffusion rank error by @divyashreepathihalli in #2208
• Simplify running KerasCV with Keras 3 by @divyashreepathihalli in #2179
• Fix dtype support for SegmentAnythingModel by @tirthasheshpatel in #2207
• Align backend/config with Keras NLP by @sampathweb in #2217
• Fix SegFormer Presets by @nkovela1 in #2222
• Update README.md by @divyashreepathihalli in #2223
• Adds Kokoro GPU Tests by @sampathweb in #2224
• Update random_cutout.py to be a subclass of VectorizedBaseImageAugmentationLayer by @sup3rgiu in #2123
• Vectorized implementation of RandomColorDegeneration and Equalization preprocessing layers by @sup3rgiu in #2214
• Fix the issue(#2195): Improve readability and comprehensibility of Stable Diffusion source by @y-vectorfield in #2197
• Fix GPU Test driver script by @sampathweb in #2228
• fix jax failing tests by @divyashreepathihalli in #2231
• Change reference of SeedGen and Serialization by @sampathweb in #2241
• Remove xlarge from continuous build by @sampathweb in #2242
• Fixes GPU Tests - JAX, PyTorch and some failures for Keras2 by @sampathweb in #2243
• Update Random to Stateless - Disables Stable Diffussion tests by @sampathweb in #2245
• Support non-tensorflow backends in KerasCV's preprocessing layers by @tirthasheshpatel in #2240
• Fix keras.engine import in predict_utils.py by @divyashreepathihalli in #2248
• Fix drop block by @divyashreepathihalli in #2250
• fix 3d preprocessing layer by @divyashreepathihalli in #2252
• Fixing conversion of box formats on the original boxes by @ariG23498 in #2213
• Add dependabot to update GitHub Actions and Python dependencies by @pnacht in #2259
• Remove cloudbuild config by @mattdangerw in #2260
• align pip_build with keras by @sampathweb in #2256
• Merge Kaggle integration to CV master branch by @nkovela1 in #2229
• Remove weight traversal output by @nkovela1 in #2275
• Align keras requirments fo torch by @sampathweb in #2272
• Modify preset kaggle handles to full path by @nkovela1 in #2279
• Bump version to stable 0.8 release by @nkovela1 in #2280

New Contributors

• @aaudevart made their first contribution in #2071
• @pnacht made their first contribution in #2075
• @mattdangerw made their first contribution in #2079
• @BjarneHerland made their first contribution in #2077
• @Philmod made their first contribution in #2109
• @AniketP04 made their first contribution in #2106
• @VBPMP made their first contribution in #2117
• @y-vectorfield made their first contribution in #2159
• @EdIzaguirre made their first contribution in #2118
• @sup3rgiu made their first contribution in #2123

Full Changelog: v0.7.0...v0.8.0