luca96 / face-clustering Goto Github PK

When I fit the model, I received this error: TypeError: Expected int64 passed to parameter 'y' of op 'Maximum', got 1e-12 of type 'float' instead. Error: Expected int64, but got 1e-12 of type 'float'.
which can be traced back to the warning TypeError: not all arguments converted during string formatting during the train generator,
is there any fix for this?

Thank you

While trying to run your pre-trained model via UL19_Clustering.ipynb notebook I get an error while executing this cell:

`model = keras.models.load_model("/content/weights-FC37-MobileNetV2-0.92.hdf5", custom_objects={'loss_max': loss_max})

model.summary()`

This is the error:

`ValueError Traceback (most recent call last)

in ()
3 return K.max(K.abs(y_pred - y_true), axis=-1)
4
----> 5 model = keras.models.load_model("/content/weights-FC37-MobileNetV2-0.92.hdf5", custom_objects={'loss_max': loss_max})
6
7 model.summary()

5 frames

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/utils/generic_utils.py in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name)
376 if obj is None:
377 raise ValueError(
--> 378 'Unknown ' + printable_module_name + ': ' + object_name)
379 # Classes passed by name are instantiated with no args, functions are
380 # returned as-is.

ValueError: Unknown loss function: cosine_proximity`

Got the same error with tf.keras instead of keras. Any ideas?

I have run github implementation of your method several times (I did full training 6 or 7 times, 12 epoches each) and I did not achive more than 80% of accurancy during training. What are "secret steps" I have to undertake to obtain 89% of accuracy in the first two epochs? :-) Are there some network tuning options that are not set on github?

Is pytorch version available for this ?

Please note the TensorFlow version in the README (I had to find out the hard way) Spoiler alert: It's not TensorFlow 2.