Here, let’s say my network is so simple such as (it’s just a minimal example)
with tf.Graph().as_default():
a = tf.placeholder(tf.float32, (None, 3), name="input")
c = a + 1
where the input placeholder has shape (None, 3).
In UFF conversion, there is a part like
uff_model = uff.from_tensorflow_frozen_model(frozen_graph_file, output_names)
parser = uffparser.create_uff_parser()
parser.register_input('Placeholder', shape, 0)
where the “register_input” requires its shape to be 3-tuple with CHW order. assuming input is always an image. In this case, I just want to use general (None, 3) shape. Is there a way for general (non-image-like) shape of input?
For information, the output is as below.
Using output node add
Converting to UFF graph
No. nodes: 11
registered input: Placeholder
TypeError: List must be length 3