Problem with SSD Mobilenet TensorRT conversion

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

HI, I am trying to convert SSD mobilenet model (ssd_mobilenet_v2_coco_2018_03_29) to an Engine file. I followed all the steps given in the page, https://github.com/AastaNV/TRT_object_detection. I get the below error,
Traceback (most recent call last):
File “main.py”, line 19, in
ctypes.CDLL("/home/einfochips/Xavier board all codes latest/TRT_object_detection-master/lib/libflattenconcat.so")
File “/usr/lib/python3.6/ctypes/init.py”, line 348, in init
self._handle = _dlopen(self._name, mode)
OSError: libnvinfer.so.5: cannot open shared object file: No such file or directory

I followed the below link

and tried to build “libflattenconcat.so”, but I am unable to build. As I checked, I am using TensorRT 7.0 version, where this file comes pre-installed.

I am stuck here for days. KIndly help me to solve this issue.

Hardware : Jetson Nano
OS: Ubuntu 18.04
Cuda : version 10.1
Tensorflow : >2
TensorRT 7.0

#3

Hi,

You don’t need to manually mount the plugin with TensorRT v7.0.
Please remove the below line and try it again.

Thanks.

#4

@AastaLLL Thanks. I removed the line18 : ctypes.CDLL(“lib/libflattenconcat.so”)

I get the below error,

2020-12-08 10:31:53.822486: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.2
Graph created successfully
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: “Input”
op: “Placeholder”
attr {
key: “dtype”
value {
type: DT_FLOAT
}
}
attr {
key: “shape”
value {
shape {
dim {
size: 1
}
dim {
size: 3
}
dim {
size: 300
}
dim {
size: 300
}
}
}
}
]
=========================================

Using output node NMS
Converting to UFF graph
Warning: No conversion function registered for layer: NMS_TRT yet.
Converting NMS as custom op: NMS_TRT
WARNING:tensorflow:From /usr/lib/python3.6/dist-packages/uff/converters/tensorflow/converter.py:226: The name tf.AttrValue is deprecated. Please use tf.compat.v1.AttrValue instead.

Warning: No conversion function registered for layer: GridAnchor_TRT yet.
Converting GridAnchor as custom op: GridAnchor_TRT
Warning: No conversion function registered for layer: FlattenConcat_TRT yet.
Converting concat_box_loc as custom op: FlattenConcat_TRT
Warning: No conversion function registered for layer: FlattenConcat_TRT yet.
Converting concat_box_conf as custom op: FlattenConcat_TRT
DEBUG [/usr/lib/python3.6/dist-packages/uff/converters/tensorflow/converter.py:143] Marking [‘NMS’] as outputs
No. nodes: 1094
UFF Output written to tmp.uff
#assertionflattenConcat.cpp,49
Aborted (core dumped)

After this I get the error saying “Python3 stopped unexpectedly”, you dont have enough memory to analyse the problem.

But $ df -h gives,

Filesystem Size Used Avail Use% Mounted on
/dev/mmcblk0p1 59G 27G 30G 48% /
none 1.8G 0 1.8G 0% /dev
tmpfs 2.0G 20M 2.0G 1% /dev/shm
tmpfs 2.0G 44M 1.9G 3% /run
tmpfs 5.0M 4.0K 5.0M 1% /run/lock
tmpfs 2.0G 0 2.0G 0% /sys/fs/cgroup
tmpfs 397M 16K 397M 1% /run/user/120
tmpfs 397M 112K 397M 1% /run/user/1000

#5

@AastaLLL @kayccc: hello team, kindly help to solve this issue. I have checked the board is at maximum performance level using command: sudo nvpmodel -q, which gives > NV Power Mode: MAXN

Unable to figure out the source of error

#6

Hi,

Sorry for the late update.

The error is independent of the device clock mode, but the configure itself.
Have you updated the configure based on your model yet?
https://github.com/AastaNV/TRT_object_detection/blob/master/config/model_ssd_mobilenet_v2_coco_2018_03_29.py

If yes, please run the tmp.uff model via trtexec with VERBOSE enabled, and share the log with us.

$ /usr/src/tensorrt/bin/trtexec --uff=tmp.uff --uffInput=Input,3,300,300 --output=NMS --verobse

Thanks.

#7

@AastaLLL: Hi, I did not do any modifications in the config, as I am running the input model as it is. Please find the output of log as you mentioned,

output_log (468.3 KB)

#8

Hi,

Sorry that the repository is published for TensorRT 5.0 and doesn’t update for a while.

For TensorRT 7.1.3 with ssd_mobilenet_v2_coco_2018_03_29 downloaded from this page.
You can run the main.py after applying the below update:

diff --git a/config/model_ssd_mobilenet_v2_coco_2018_03_29.py b/config/model_ssd_mobilenet_v2_coco_2018_03_29.py
index 3c9f3b8..4046ce5 100644
--- a/config/model_ssd_mobilenet_v2_coco_2018_03_29.py
+++ b/config/model_ssd_mobilenet_v2_coco_2018_03_29.py
@@ -55,11 +55,13 @@ def add_plugin(graph):
     concat_box_loc = gs.create_plugin_node(
         "concat_box_loc",
         op="FlattenConcat_TRT",
+        axis=1
     )

     concat_box_conf = gs.create_plugin_node(
         "concat_box_conf",
         op="FlattenConcat_TRT",
+        axis=1
     )

     namespace_plugin_map = {
diff --git a/main.py b/main.py
index 09537e5..a002c33 100644
--- a/main.py
+++ b/main.py
@@ -15,7 +15,6 @@ import graphsurgeon as gs
 #from config import model_ssd_mobilenet_v1_coco_2018_01_28 as model
 from config import model_ssd_mobilenet_v2_coco_2018_03_29 as model

-ctypes.CDLL("lib/libflattenconcat.so")
 COCO_LABELS = coco.COCO_CLASSES_LIST


@@ -28,7 +27,7 @@ runtime = trt.Runtime(TRT_LOGGER)
 # compile model into TensorRT
 if not os.path.isfile(model.TRTbin):
     dynamic_graph = model.add_plugin(gs.DynamicGraph(model.path))
-    uff_model = uff.from_tensorflow(dynamic_graph.as_graph_def(), model.output_name, output_filename='tmp.uff')
+    uff_model = uff.from_tensorflow(dynamic_graph.as_graph_def(), model.output_name, output_filename='tmp.uff',text=True)

     with trt.Builder(TRT_LOGGER) as builder, builder.create_network() as network, trt.UffParser() as parser:
         builder.max_workspace_size = 1 << 28

We have checked this and confirm it works in our environment.
Please share your result with us and sorry for the inconvenience.

Thanks.