tf.keras
allows model weights to be
saved in two formats: HDF5 and TensorFlow format. Currently only TensorFlow format is
supported by the tool. If the model weights have been saved in HDF5, then you have to
convert it to TensorFlow format.
import tensorflow as tf
tf.keras.backend.set_learning_phase(0)
model = tf.keras.applications.ResNet50(weights="imagenet",
include_top=True,
input_tensor=None,
input_shape=None,
pooling=None,
classes=1000)
model.save_weights("model.ckpt", save_format='tf')
The converted checkpoint files look like this:
model.ckpt.data-00000-of-00001
model.ckpt.data-00001-of-00002
model.ckpt.index