StridedSlice only support Constant nodes as input for now

When I parse a UFF converted from a tensorflow model, the error occurs:
ERROR: UFFParser: Parser error: output/strided_slice_1: StridedSlice only support Constant nodes as input for now.
What does it mean and how can I solve it?

Hi,

Could you share your network with us?

For strided_slice, UFF parser only supports constant input.
https://www.tensorflow.org/api_docs/python/tf/strided_slice

input_ needs to be a constant layer rather than variable.

Thanks.

Hi AastaLLL:
The part of the graph is as below:
predictions = tf.reshape(predictions, [1, side_, side_, num, -1])
predictions = tf.constant(predictions[0])
pre_obj = predictions[:, :, :, :, 4]
pre_obj = tf.sigmoid(pre_obj)
pre_obj = tf.reshape(pre_obj, [1, side_, side_, num, 1])
pre_coord = predictions[:, :, :, :, 0:4]
pre_class = predictions[:, :, : ,:, 5:deep_]
pre_class = tf.reshape(pre_class, [-1, classes_])
pre_class = tf.nn.softmax(pre_class)
pre_class = tf.reshape(pre_class, [1, side_, side_, num, classes_])
pre_out = tf.concat([pre_coord, pre_obj, pre_class], 4)

the problem is caused by pre_obj = predictions[:, :, :, :, 4]. How can I make the parser to parse it correctly?.

Hi,

Slice layer is not supported.
Sorry for the inconvenience.

Thanks.

Hi Aastall:

Can I use plugin to implement these operateion?

Hi,

Yes, since it is a weight-free layer.
You can convert it to UFF format without this layer first, and implement it with Plugin API.

Thanks.