Error on running detect_objects_webcam.py

[alfr137]

Machine settings: Ubuntu 18.04.3 LTS, Nvidia Quadro RTX 6000/PCIe/SSE2, NVIDIA-SMI 430.26 Driver Version: 430.26 , CUDA Version: 10.2 ;

I managed to setup the environment using the ./setup_environment.sh but when i try to run the inference on webcam using ./setup_environment.sh this error occurs:

"SSD_Model/detect_objects_webcam.py", line 12, in
import utils.inference as inference_utils # TRT/TF inference wrappers
File "/mnt/SSD_Model/utils/inference.py", line 60, in
import utils.engine as engine_utils # TRT Engine creation/save/load utils
File "/mnt/SSD_Model/utils/engine.py", line 11, in
from utils.model import ModelData
File "/mnt/SSD_Model/utils/model.py", line 9, in
import graphsurgeon as gs
File "/usr/lib/python3.5/dist-packages/graphsurgeon/init.py", line 9, in
from graphsurgeon.StaticGraph import *
File "/usr/lib/python3.5/dist-packages/graphsurgeon/StaticGraph.py", line 7, in
from graphsurgeon._utils import _regex_list_contains_string, _generate_iterable_for_search, _clean_input_name
File "/usr/lib/python3.5/dist-packages/graphsurgeon/_utils.py", line 2, in
from tensorflow import NodeDef
ImportError: cannot import name 'NodeDef'

[zarar7576]

It's probably because of a tensorflow version issue because python is trying to import NodeDef and can't find it:

"from tensorflow import NodeDef
ImportError: cannot import name 'NodeDef'"

Which version of TensorFlow are you using?

If you are on 2.0 then try downgrading to 1.14.0
or else try upgrading to 1.14.0
I tried it in TesorFlow version 1.14.0 and 1.12.2 and it worked

[cassini12]

Anyone get this working? I am at the same exact spot.. same error

cassini@cassini51:$ python --version
Python 2.7.15+
cassini@cassini51:$ python3 --version
Python 3.6.8

cassini@cassini51:~$ pip3 list |grep tensorflow
DEPRECATION: The default format will switch to columns in the future. You can use --format=(legacy|columns) (or define a format=(legacy|columns) in your pip.conf under the [list] section) to disable this warning.
tensorflow (1.15.0)
tensorflow-estimator (1.13.0)
tensorflow-gpu (1.14.0)

[zarar7576]

What TensorFlow version are you using in docker?
I had to install TensorFlow 1.14.0 for it to run

pip install tensorflow==1.14.0
pip install tensorflow-gpu==1.14.0

[cassini12]

Just ran both of those commands, I honestly am not sure. I am new to all this. Was trying to follow along with the medium article just to familiarize myself with it. At first my issue was the setup_env but Now I feel I got that squared away and I am facing the NoDef issue

[cassini12]

cassini@cassini51:~/Object-Detection/object-detection-tensorrt-example$ sudo ./setup_environment.sh
Setting envivonment variables for the webcam
non-network local connections being added to access control list
VOCdevkit has already been downloaded
Dockerfile has already been built
Starting docker container

=====================
== NVIDIA TensorRT ==

NVIDIA Release 19.05 (build 6392482)

NVIDIA TensorRT 5.1.5 (c) 2016-2019, NVIDIA CORPORATION. All rights reserved.
Container image (c) 2019, NVIDIA CORPORATION. All rights reserved.

https://developer.nvidia.com/tensorrt

To install Python sample dependencies, run /opt/tensorrt/python/python_setup.sh

root@81666cad449d:/mnt# python SSD_Model/detect_objects_webcam.py
Traceback (most recent call last):
File "SSD_Model/detect_objects_webcam.py", line 12, in
import utils.inference as inference_utils # TRT/TF inference wrappers
File "/mnt/SSD_Model/utils/inference.py", line 60, in
import utils.engine as engine_utils # TRT Engine creation/save/load utils
File "/mnt/SSD_Model/utils/engine.py", line 11, in
from utils.model import ModelData
File "/mnt/SSD_Model/utils/model.py", line 9, in
import graphsurgeon as gs
File "/usr/lib/python3.5/dist-packages/graphsurgeon/init.py", line 9, in
from graphsurgeon.StaticGraph import *
File "/usr/lib/python3.5/dist-packages/graphsurgeon/StaticGraph.py", line 7, in
from graphsurgeon._utils import _regex_list_contains_string, _generate_iterable_for_search, _clean_input_name
File "/usr/lib/python3.5/dist-packages/graphsurgeon/_utils.py", line 2, in
from tensorflow import NodeDef
ImportError: cannot import name 'NodeDef'
root@81666cad449d:/mnt#

[zarar7576]

I had the same issue but after running

python_setup.sh

I downgraded the TensorFlow version inside the container by using pip

pip install tensorflow==1.14.0
pip install tensorflow-gpu==1.14.0

and it worked beautifully I even managed to use a version of Faster RCNN on this, I hope you resolve this issue TRT is a great tool

[cassini12]

Update: pip install tensorflow==1.14.0
Running this worked!
originally I only had done the GPU to 14 but it seems I needed both. Thank you

follow up, am I able to "modify" this object detection to include pictures of people/things I have? and train it to recognize distinctly?

[zarar7576]

You can you will need to modify the OpenCV input code ill share something tomorrow that might help you
It takes a bit of work but once you get it, it's pretty easy and you can modify it for any object you want

[cassini12]

Awesome, I look forward to that! Thank you for all your help and support

[cassini12]

If anyone knows, After running it once, I have to RM the .buf before it will run again. Is this normal?
/mnt/SSD_Model/utils/../workspace/engines/FLOAT/engine_bs_1.buf