Add support for PyTorch and Jax to KerasCV #1925
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#1903) * Port preprocessing layers and bounding box utils to Keras Core * Fix pip install * Another git clone fix * Fix for nested scopes * Use PAT for git clone * PAT structure * Update actions.yml * Update actions.yml * Update actions.yml * Int64 for jax * Update utils * Remove commented-out code * ANY * No any typehints * Deploy key for install * Remove prints * SSH key as github action variable * New ssh key approach * Review comments * Fancy pytest markers * operations -> ops plus some test marks * Add torch hax * Fix TF marker * Format * Review comments * rename tf_only to tf_keras_only * Try installing keras core with pip_build * s/_/- * namex * Namex fix + torch * Rich * requirements.txt * I promise I have used a computer before * Back to private API for now * Fix path for backend * Back to namex install * Use correct policy API for mixed precision * Reverse aliases * Rely on keras-core for validation of backend * Copy the homework of matt * Newline
* Port backbones to Keras Core * Fix YOLOV8 presets * Fix tests * Fix backbones * utils -> saving * YOLOV8 presets -- again * saving for csp * Test fixes * Squeeze_excite Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com>
* Port losses to Keras Core * Saving * Fix focal * Update CenterNetBoxLoss Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com>
* Port ImageClassifier to Keras Core * Backbone property
* Port OD layers to Keras Core * Add multi-backend NMS * Use pytest fixture for skippage * Nice asserts for tf.keras-only components * Review comments Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com>
* Port RetinaNet to Keras Core * Add defensive tuple casting * Remove Torch one_hot workaround * Cast to int
* Use torch-native NMS where possible * Implement native NMS for Torch * Remove prints * valid_det is an int, foo * fix num dets * Comment
* Add keras-core as a dependency * Install keras core from PyPi on CI * Fix tf.keras CI
* Port YOLOV8 to Keras Core' * Update test utils * Mark backbone tests XL' --------- Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com>
|
A note for reviewers -- this code has all already been reviewed via PRs to the |
|
/gcbrun |
|
/gcbrun |
1 similar comment
|
/gcbrun |
|
/gcbrun |
|
/gcbrun |
|
/gcbrun |
|
/gcbrun |
fchollet
approved these changes
Jul 11, 2023
ghost
pushed a commit
to y-vectorfield/keras-cv
that referenced
this pull request
Nov 16, 2023
* Port preprocessing layers and bounding box utils to Keras Core [try 2] (keras-team#1903) * Port preprocessing layers and bounding box utils to Keras Core * Fix pip install * Another git clone fix * Fix for nested scopes * Use PAT for git clone * PAT structure * Update actions.yml * Update actions.yml * Update actions.yml * Int64 for jax * Update utils * Remove commented-out code * ANY * No any typehints * Deploy key for install * Remove prints * SSH key as github action variable * New ssh key approach * Review comments * Fancy pytest markers * operations -> ops plus some test marks * Add torch hax * Fix TF marker * Format * Review comments * rename tf_only to tf_keras_only * Try installing keras core with pip_build * s/_/- * namex * Namex fix + torch * Rich * requirements.txt * I promise I have used a computer before * Back to private API for now * Fix path for backend * Back to namex install * Use correct policy API for mixed precision * Reverse aliases * Rely on keras-core for validation of backend * Copy the homework of matt * Newline * Port backbones to Keras Core (keras-team#1906) * Port backbones to Keras Core * Fix YOLOV8 presets * Fix tests * Fix backbones * utils -> saving * YOLOV8 presets -- again * saving for csp * Test fixes * Squeeze_excite Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com> * Port losses to Keras Core (keras-team#1905) * Port losses to Keras Core * Saving * Fix focal * Update CenterNetBoxLoss Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com> * Port ImageClassifier to Keras Core (keras-team#1908) * Port ImageClassifier to Keras Core * Backbone property * Port object detection layers to Keras Core (keras-team#1907) * Port OD layers to Keras Core * Add multi-backend NMS * Use pytest fixture for skippage * Nice asserts for tf.keras-only components * Review comments Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com> * Backbone property fix (keras-team#1909) * Port RetinaNet to Keras Core (keras-team#1912) * Port RetinaNet to Keras Core * Add defensive tuple casting * Remove Torch one_hot workaround * Cast to int * Port PyCOCOCallback to Keras Core (keras-team#1913) * Fix default training value for SqueezeExcite (keras-team#1917) * Update to_numpy to use Keras Core ops (keras-team#1916) * Implement native NMS for PyTorch backend (keras-team#1918) * Use torch-native NMS where possible * Implement native NMS for Torch * Remove prints * valid_det is an int, foo * fix num dets * Comment * Bump version to 0.6.0 for Keras Core release (keras-team#1919) * Add keras-core as a dependency (keras-team#1922) * Add keras-core as a dependency * Install keras core from PyPi on CI * Fix tf.keras CI * Port YOLOV8 to Keras Core' (keras-team#1920) * Port YOLOV8 to Keras Core' * Update test utils * Mark backbone tests XL' --------- Co-authored-by: Tirth Patel <tirthasheshpatel@gmail.com> * Fix saving namespace for TF < 2.12 * Fix penalty reduced focal loss * Run GPU tests on tf 2.13, require 2.12 * Run cpu tests on tf 2.13, require 2.13 * Fix ABI and FasterRCNN tests --------- Co-authored-by: Ian Stenbit <3072903+ianstenbit@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR adds support for Keras Core, a new multi-backend implementation for Keras with support for TensorFlow, Jax, and PyTorch. Several backbones and layers have been ported to keras-core along with two object detection models, RetinaNet and YOLOV8.
Backbones ported:
ResNetV2Backbone,EfficientNetV2Backbone, andImageClassifierto Keras Core ianstenbit/keras-cv#2)ResNetV2Backbone,EfficientNetV2Backbone, andImageClassifierto Keras Core ianstenbit/keras-cv#2)ResNetV2Backbone,EfficientNetV2Backbone, andImageClassifierto Keras Core ianstenbit/keras-cv#2)Other components ported:
RetinaNet PR: ianstenbit#13
YOLOV8 PR: ianstenbit#8, #1920
Who can review?
@ianstenbit @jbischof @fchollet