Expecting KerasTensor - Got: <BatchDataset shapes: ...>

user_3060692 0
Viewed 573

Beginners question:

I collect images (128x128x3, batch_size=32) via tf.keras.preprocessing.image_dataset_from_directory and try to process these images via DenseNet121.

But it ends up in:

Found unexpected instance while processing input tensors for keras functional model. Expecting KerasTensor which is from tf.keras.Input() or output from keras layer call(). Got: <BatchDataset shapes: ((None, 128, 128, 3), (None,)), types: (tf.float32, tf.int32)>

How do I have to shape my Dataset that it is compatible with DenseNet?

tensorflow-datasets tensorflow2.0 keras tensorflow python
2 Answers

Densenet with Keras required input to be in a specific shape and type. Before you pass your inputs into your model you have to run the images through tf.keras.applications.densenet.preprocess_input to get your inputs into the correct shape and format.

Note that tf.keras.applications.densenet.preprocess_input expects input to be a numpy.array or a tf.Tensor that is 3D or 4D (https://www.tensorflow.org/api_docs/python/tf/keras/applications/densenet/preprocess_input).

Also, tf.keras.preprocessing.image_dataset_from_directory returns a tf.data.Dataset which may need (I am not sure if the preprocessing function accepts it) processing to get it into a tensor or numpy array.

Hope this helped!

user_180227590

You do not need to use tf.data.Dataset.from_tensor_slices at all. Try something like this:

import tensorflow as tf
import pathlib

dataset_url = "https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz"
data_dir = tf.keras.utils.get_file('flower_photos', origin=dataset_url, untar=True)
data_dir = pathlib.Path(data_dir)

batch_size = 32

train_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="training",
  seed=123,
  image_size=(128, 128),
  batch_size=batch_size)

train_ds = train_ds.map(lambda x, y: (tf.keras.applications.densenet.preprocess_input(x), y))

dense_net = tf.keras.applications.DenseNet121(include_top=False, weights="imagenet",input_shape=(128, 128, 3))

inputs = tf.keras.layers.Input((128, 128, 3))
x = dense_net(inputs)
x = tf.keras.layers.GlobalMaxPool2D()(x)
outputs = tf.keras.layers.Dense(5)(x)
model = tf.keras.Model(inputs, outputs)

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
history = model.fit(train_ds, epochs=5)
user_96578610
Related