-
**Short Description**
One major issue is that the model will perform poorly on new, unseen data, even if trained well on the existing data, because of overfitting.
**Papers**
https://arxiv.o…
-
To align KerasCV with the KerasNLP API, we're migrating our backbone models to subclass keras_cv.models.Backbone.
This allows us to have a more structured, modular API for high-level task models, inc…
-
Hey folks,
I encountered something strange in `keras-cv-0.4.2`. Long story short, I believe there are something wrong with the provided stable diffusion model's VQVAE image encoder/decoder or rela…
-
This should follow the example of #1412 and #1401, including
- Docstring updates
- Colab demonstrating that weight loading works (if pre-trained weights exist) and numerics are identical
-
Hello @divamgupta ! As you know, we ported the image encoder in https://github.com/keras-team/keras-cv from your library! We've supported inpainting via a method on StableDiffusion, and we're seeing…
-
I find that current implementation of `VectorizedBaseImageAugmentationLayer` fails to correctly process unbatched `bounding_boxes`
Here is a standalone script:
```python
import tensorflow as tf…
-
To align KerasCV with the KerasNLP API, we're migrating our backbone models to subclass keras_cv.models.Backbone.
This allows us to have a more structured, modular API for high-level task models, inc…
-
Currently this breaks when operating on ragged images. This can be see in the `examples/layers/random_shear_demo.py` demo.
More info available here: https://github.com/keras-team/keras-cv/pull/153…
-
When creating a classifier via `from_preset()` or the default constructor, we should require `num_classes` to be passed. This will be consistent with recent updates to keras-cv.
```python
model = …
-
Paper: https://arxiv.org/abs/2106.09685
```
LoRA is a finetuning method which freezes the pre-trained model weights and injects trainable rank decomposition matrices into each layer of the Transfo…