ValueError: Unrecognized keyword arguments passed to Dense

[PythoneerSamurai]

I've tried to make a keras_ocr.pipeline.Pipeline() object both on my local machine and cloud notebook, and I'm getting the error mentioned in the title of this issue. I have tensorflow > 2.0.0, as mentioned on the GitHub page. How to fix this? Another point to be noted is that I opened a very old notebook from a youtuber, it was using tensorflow 2.6.4, and keras_ocr was working properly there. However, I cannot install that version of tensorflow as pip does not recognize it, probably because it is not compatible with newer versions of python.

[vigneshwarvenkat]

Had the same issue, but the fix specified on #252 works for me.

tensorflow==2.13.0
keras==2.13.1
keras-ocr==0.9.3

[PythoneerSamurai]

Had the same issue, but the fix specified on #252 works for me.

tensorflow==2.13.0 keras==2.13.1 keras-ocr==0.9.

Thankyou, I installed your versions of tensorflow, keras, and keras-ocr. Now my code is running.

[Deependrashukla]

had the same issue and fixed it by installing this version in colab.
!pip install tensorflow==2.15

[smathieson]

This is useful, but are there any plans for updating and supporting keras3?

[t-wolfeadam]

FYI since TF 2.15 is not compatible with python 3.12+, what I did to fix it was change L277:282 in recognition.py to:

locnet_y = keras.layers.Dense(
            6,
            kernel_initializer=tf.constant_initializer(np.zeros((64, 6), dtype="float32")),
            bias_initializer=tf.constant_initializer(np.array([[1, 0, 0], [0, 1, 0]], dtype="float32").flatten()),
        )(locnet_y)