Port the ImageClassifier guide to keras_core
#608
Conversation
| ## Multi-Backend Support | ||
|
|
||
| Keras-CV's `ImageClassifier` model supports several backends like Jax, PyTorch, and TensorFlow with the help of `keras_core`. To enable multi-backend support in Keras-CV, set the `KERAS_CV_MULTI_BACKEND` environment variable. We can then switch between different backends by setting the `KERAS_BACKEND` environment variable. Currently, `tensorflow`, `jax`, and `torch` are supported. | ||
| KerasCV's `ImageClassifier` model supports several backends like Jax, PyTorch, |
| and TensorFlow with the help of `keras_core`. To enable multi-backend support | ||
| in KerasCV, set the `KERAS_CV_MULTI_BACKEND` environment variable. We can | ||
| then switch between different backends by setting the `KERAS_BACKEND` | ||
| environment variable. Currently, `tensorflow`, `jax`, and `torch` are |
There was a problem hiding this comment.
Use backticks + double quotes for strings
tirthasheshpatel
changed the title
ImageClassifier guide to keras_core"ImageClassifier guide to keras_core
|
|
||
| """Now let's apply our final augmenter to the training data:""" | ||
|
|
||
| augmenter = tf.keras.Sequential(augmenters) |
There was a problem hiding this comment.
There should be no tf.keras at all
There was a problem hiding this comment.
Let's use the Augmenter which we're adding back to KerasCV in keras-team/keras-cv#1978
|
|
||
|
|
||
| def preprocess_inputs(image, label): | ||
| image = tf.cast(image, tf.float32) |
There was a problem hiding this comment.
Since the example has quite a bit of TF-only APIs...
- We could reduce or remove the TF dependency and keep it in the backend-agnostic folder.
- We could move it to the TF-only folder.
There was a problem hiding this comment.
The TF components are just for preprocessing which for now has to live in tf.data. I'm fine with moving it to TF-only, but I would expect this to work fine with all backends since this is constrained to the tf.data parts of the example
There was a problem hiding this comment.
+1 for what @ianstenbit said. I am OK with moving this under TF-only though, if you prefer @fchollet.
There was a problem hiding this comment.
I think at minima we should:
- drop the dependency on
tf.keras.Sequentialand just use a list or Augmenter - drop the dependency on
tf.one_hot. You can use aCategoryEncodinglayer for that - remove the tf.cast to float (unnecessary)
- remove
import tensorflowand dofrom tensorflow import data as tf_data
There was a problem hiding this comment.
I was trying your suggestions out @fchollet and seems like we would still need the tf import for one hot encoding since CategoryEncoding doesn't accept float32 labels (so, even if we use it, we'd still need to cast labels to integers).
Other than that, everything else works. Let me know if this is OK from your side!
|
Can you also open a PR for this on keras-team/keras-io? |
|
Sure, I still had one change in flight but I'll submit a follow-up! I will also submit a PR on keras.io. Thanks for the reviews @fchollet @ianstenbit! |
Continuing #273.