Uffparser: validator error: ones_like: unsupported operation _Fill

Hi, I tried to use tensorRT speed up my network. I parse .forzen_model.pb to .uff. It got the error as follow.

uffparser: validator error: ones_like: unsupported operation _Fill.

I use the tf operator tf.ones_like() and tf.zeros_like() in my network.

How can I fix this?

Hi,

We do have IFillLayerin TensorRT API but not in the uff parser:
https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#fill-layer
Would you mind to try .pb -> .onnx -> .trt to see if works?

Thanks.

I tried convert pb to onnx. While I got another issue when doing running (c++).

[09/02/2020-17:34:24] [W] [TRT] onnx2trt_utils.cpp:220: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
ERROR: builtin_op_importers.cpp:1554 In function importIf:
[8] Assertion failed: cond.is_weights() && cond.weights().count() == 1 && “If condition must be a initializer!”

Do we have any ways to avoid this error.

done

Hi,

Do you fix this issue already?
Would you mind to share more information/status with us?

Thanks.

The main reason is that the network contains the unknown shape tensor.
I use tf.placeholder(tf.float, name=“"), but i did not specify the shape. It works fine with
tf.placeholder(tf.float, [none, 1],name="
”)