Segmentation fault (core dumped) uff model loading

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Hi,

I converted my frozen .pd posenet model in uff format. Then, when I want to create the builder, network, and parser using python with the following instructions:
with builder = trt.Builder(TRT_LOGGER) as builder, builder.create_network() as network, trt.UffParser() as parser:
parser.register_input(“Placeholder”, (1, 28, 28))
parser.register_output(“fc2/Relu”)
parser.parse(model_file, network)

I get the following error: Segmentation fault (core dumped)

Is my model supported by tensorrt or did I make another mistakes ?

Thanks

1 Like
3

Hi,

Usually, segmentation fault is caused by memory-related error.
Would you mind to share a complete log with us so we can check it for you?

Thanks.

4

Thanks for the reply,

The only error message I get is “Segmentation fault (core dumped)” and the following file is generated: core.1808 each time I run my python script.

I also tried with python sample example to run my model and get the following error:
[TensorRT] ERROR: UffParser: Could not open models/posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.uff
[TensorRT] ERROR: Network must have at least one output
[TensorRT] ERROR: Network validation failed.

Then I tried
“python convert_to_uff.py posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.pb -O MNS” but resulted to a fail during convertion:

raise UffException(str(name) + " was not found in the graph. Please use the -l option to list nodes in the graph.")
uff.model.exceptions.UffException: NMS was not found in the graph. Please use the -l option to list nodes in the graph.

I have no idea what kind of model the converter is waiting for running in inference next.

5

Hi,

The error complains about no .uff is found.
Would you mind to check if you feed the file correctly?

ERROR: UffParser: Could not open models/posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.uff

Thanks.

6

It founds my model and now I’m getting the following error:

Order size is not matching the number dimensions of TensorRT
Building TensorRT engine, this may take a few minutes…
[TensorRT] ERROR: Network must have at least one output
[TensorRT] ERROR: Network validation failed.

During conversion I get the following message:

Loading posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.pb
NOTE: UFF has been tested with TensorFlow 1.15.0.
WARNING: The version of TensorFlow installed on this system is not guaranteed to work with UFF.
UFF Version 0.6.9
=== Automatically deduced input nodes ===
[name: “sub_2”
op: “Placeholder”
attr {
key: “dtype”
value {
type: DT_FLOAT
}
}
attr {
key: “shape”
value {
shape {
dim {
size: 1
}
dim {
size: 257
}
dim {
size: 257
}
dim {
size: 3
}
}
}
}
]

=== Automatically deduced output nodes ===
[name: “MobilenetV1/heatmap_2/BiasAdd”
op: “Add”
input: “MobilenetV1/heatmap_2/BiasAdd/conv2d”
input: “MobilenetV1/heatmap_2/BiasAdd/conv_bias/read”
attr {
key: “T”
value {
type: DT_FLOAT
}
}
, name: “MobilenetV1/offset_2/BiasAdd”
op: “Add”
input: “MobilenetV1/offset_2/BiasAdd/conv2d”
input: “MobilenetV1/offset_2/BiasAdd/conv_bias/read”
attr {
key: “T”
value {
type: DT_FLOAT
}
}
, name: “MobilenetV1/displacement_bwd_2/BiasAdd”
op: “Add”
input: “MobilenetV1/displacement_bwd_2/BiasAdd/conv2d”
input: “MobilenetV1/displacement_bwd_2/BiasAdd/conv_bias/read”
attr {
key: “T”
value {
type: DT_FLOAT
}
}
, name: “MobilenetV1/displacement_fwd_2/BiasAdd”
op: “Add”
input: “MobilenetV1/displacement_fwd_2/BiasAdd/conv2d”
input: “MobilenetV1/displacement_fwd_2/BiasAdd/conv_bias/read”
attr {
key: “T”
value {
type: DT_FLOAT
}
}
]

Using output node MobilenetV1/heatmap_2/BiasAdd
Using output node MobilenetV1/offset_2/BiasAdd
Using output node MobilenetV1/displacement_bwd_2/BiasAdd
Using output node MobilenetV1/displacement_fwd_2/BiasAdd
Converting to UFF graph
DEBUG […/…/uff/converters/tensorflow/converter.py:143] Marking [‘MobilenetV1/heatmap_2/BiasAdd’, ‘MobilenetV1/offset_2/BiasAdd’, ‘MobilenetV1/displacement_bwd_2/BiasAdd’, ‘MobilenetV1/displacement_fwd_2/BiasAdd’] as outputs
No. nodes: 156
UFF Output written to posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.uff

I have no idea where the problem comes from because the outputs are well detected like heatmap and offset outputs from the posenet model.

Before building TensorRT engine: the command line parser.parse(uff_model_path, network) results as “False”

Script:with trt.Builder(TRT_LOGGER) as builder, builder.create_network() as network, trt.UffParser() as parser:
parser.register_input(“Placeholder”, (3, 257, 257))
parser.register_output(“fc2/Relu”)
parser.parse(uff_model_path, network)

7

Hi,

It looks like the conversion between .pb->.uff is good but .uff->.engine is not working.
Is our understanding correct?

According to the code you shared, the output layer name is not synchronized.
Would you mind to update it first?

parser.register_output("MobilenetV1/heatmap_2/BiasAdd")
parser.register_output("MobilenetV1/offset_2/BiasAdd")
parser.register_output("MobilenetV1/displacement_bwd_2/BiasAdd")
parser.register_output("MobilenetV1/displacement_fwd_2/BiasAdd")

Thanks.

9

Hi,

I tried with the good names output but still exactly the same error:

[TensorRT] ERROR: UffParser: Parser error: MobilenetV1/Conv2d_0/Relu6/conv2d: Order size is not matching the number dimensions of TensorRT
Building TensorRT engine, this may take a few minutes…
[TensorRT] ERROR: Network must have at least one output
[TensorRT] ERROR: Network validation failed.

I saw that some operations are not supported by Tenssort, so need to use TF_TRT instead.