AttributeError occured

[Hiroaki-K4]

Hi!
Thank you for your work.
I tried: roslaunch trt_yolo_ros demo.launch

Then such an error occurred. If you know a solution, please let me know.

... logging to /home/nvidia/.ros/log/7194ed44-ecee-11ea-b2e1-2a67aa1ada95/roslaunch-nvidia-17905.log
Checking log directory for disk usage. This may take a while.
Press Ctrl-C to interrupt
Done checking log file disk usage. Usage is <1GB.

started roslaunch server http://nvidia:34107/

SUMMARY
========

PARAMETERS
 * /rosdistro: melodic
 * /rosversion: 1.14.6
 * /trt_yolo_ros/config_path: /home/nvidia/catk...
 * /trt_yolo_ros/cuda_device: 0
 * /trt_yolo_ros/label_filename: coco_labels.txt
 * /trt_yolo_ros/nms_threshold: 0.3
 * /trt_yolo_ros/num_cam: 1
 * /trt_yolo_ros/obj_threshold: 0.7
 * /trt_yolo_ros/postprocessor_cfg: yolo_postprocess_...
 * /trt_yolo_ros/publish_image: True
 * /trt_yolo_ros/publish_rate: 10
 * /trt_yolo_ros/publisher/bounding_boxes/topic: /trt_yolo_ros/bou...
 * /trt_yolo_ros/publisher/image/topic: /trt_yolo_ros/bou...
 * /trt_yolo_ros/subscriber/image0/topic: /camera/color/ima...
 * /trt_yolo_ros/weights_path: /home/nvidia/catk...
 * /trt_yolo_ros/yolo_type: yolov3-416

NODES
  /
    trt_yolo_ros (trt_yolo_ros/node.py)

auto-starting new master
process[master]: started with pid [17916]
ROS_MASTER_URI=http://localhost:11311

setting /run_id to 7194ed44-ecee-11ea-b2e1-2a67aa1ada95
process[rosout-1]: started with pid [17927]
started core service [/rosout]
process[trt_yolo_ros-2]: started with pid [17934]
[INFO] [1599031947.810237]: [trt_yolo_ros] starting the node
Building new trt engine file
Traceback (most recent call last):
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/scripts/node.py", line 11, in <module>
    network = YOLORos()
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/src/trt_yolo_ros/trt_yolo_ros.py", line 27, in __init__
    show_image=self.publish_image,
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/src/trt_yolo/detector.py", line 60, in __init__
    self.engine = self.get_engine(weights_path, config_path, yolo_type)
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/src/trt_yolo/detector.py", line 109, in get_engine
    self.build_trt_from_onnx(weights_path, configs_path, yolo_type)
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/src/trt_yolo/detector.py", line 125, in build_trt_from_onnx
    with trt.Builder(self.trt_logger) as builder:
AttributeError: __exit__
[trt_yolo_ros-2] process has died [pid 17934, exit code 1, cmd /home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/scripts/node.py __name:=trt_yolo_ros __log:=/home/nvidia/.ros/log/7194ed44-ecee-11ea-b2e1-2a67aa1ada95/trt_yolo_ros-2.log].
log file: /home/nvidia/.ros/log/7194ed44-ecee-11ea-b2e1-2a67aa1ada95/trt_yolo_ros-2*.log

Thanks!

[kumuji]

Can you specify your environment parameters?
Python, tensorrt versions

[Hiroaki-K4]

ok.
python 2.7
tensorrt 6.0.1

[kumuji]

ok.
python 2.7
tensorrt 6.0.1

Versions look fine, I have almost same.
It could be the problem that onnx engine was generated wrong. Can you try removing contents of weights folder in trt_yolo and rerun the node?

[Hiroaki-K4]

I deleted the contents of the weights folder and ran it again, but got the same error.

[kumuji]

Hmm, it seems close to this issue.
Which python minor(2.7.17) version are you using? Could it be related to that?

[Hiroaki-K4]

I solved it.
I changed
with trt.Builder(self.trt_logger) as builder:
to
builder = trt.Builder(self.trt_logger)
I also changed the with statement in several places.
But a new error occurred.

SUMMARY
========

PARAMETERS
 * /rosdistro: melodic
 * /rosversion: 1.14.6
 * /trt_yolo_ros/config_path: /home/nvidia/catk...
 * /trt_yolo_ros/cuda_device: 0
 * /trt_yolo_ros/label_filename: coco_labels.txt
 * /trt_yolo_ros/nms_threshold: 0.3
 * /trt_yolo_ros/num_cam: 1
 * /trt_yolo_ros/obj_threshold: 0.7
 * /trt_yolo_ros/postprocessor_cfg: yolo_postprocess_...
 * /trt_yolo_ros/publish_image: True
 * /trt_yolo_ros/publish_rate: 10
 * /trt_yolo_ros/publisher/bounding_boxes/topic: /trt_yolo_ros/bou...
 * /trt_yolo_ros/publisher/image/topic: /trt_yolo_ros/bou...
 * /trt_yolo_ros/subscriber/image0/topic: /camera/color/ima...
 * /trt_yolo_ros/weights_path: /home/nvidia/catk...
 * /trt_yolo_ros/yolo_type: yolov3-416

NODES
  /
    trt_yolo_ros (trt_yolo_ros/node.py)

auto-starting new master
process[master]: started with pid [27898]
ROS_MASTER_URI=http://localhost:11311

setting /run_id to 92fb3db4-ee77-11ea-a597-04ab18464945
WARNING: Package name "efficientNet_proc" does not follow the naming conventions. It should start with a lower case letter and only contain lower case letters, digits, underscores, and dashes.
WARNING: Package name "pyTorch_proc" does not follow the naming conventions. It should start with a lower case letter and only contain lower case letters, digits, underscores, and dashes.
process[rosout-1]: started with pid [27909]
started core service [/rosout]
process[trt_yolo_ros-2]: started with pid [27916]
[INFO] [1599200797.075055]: [trt_yolo_ros] starting the node
Reading engine from file /home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/weights/yolov3-416.trt
Traceback (most recent call last):
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/scripts/node.py", line 11, in <module>
    network = YOLORos()
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/src/trt_yolo_ros/trt_yolo_ros.py", line 27, in __init__
    show_image=self.publish_image,
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/src/trt_yolo/detector.py", line 61, in __init__
    self.inputs, self.outputs, self.bindings, self.stream = self._allocate_buffers()
  File "/home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo/src/trt_yolo/detector.py", line 181, in _allocate_buffers
    trt.volume(self.engine.get_binding_shape(binding))
AttributeError: module 'tensorrt' has no attribute 'volume'
[trt_yolo_ros-2] process has died [pid 27916, exit code 1, cmd /home/nvidia/catkin_ws/src/trt_yolo_ros/trt_yolo_ros/scripts/node.py __name:=trt_yolo_ros __log:=/home/nvidia/.ros/log/92fb3db4-ee77-11ea-a597-04ab18464945/trt_yolo_ros-2.log].
log file: /home/nvidia/.ros/log/92fb3db4-ee77-11ea-a597-04ab18464945/trt_yolo_ros-2*.log

Is the version of tensorrt not right?

[kumuji]

Is the version of tensorrt not right?

I use tensorrt 6.0.1.5 for cuda 10 with python 2.7.17, and works fine.
Looks like builder for some reason does not return the context.

[Hiroaki-K4]

ok.
Is there a way to check if the builder is working properly?

[kumuji]

I think you can check examples in $TENSORRT_PATH/samples/python/yolov3_onnx, my repo is based on that example. Can you check if it works for you?

[Hiroaki-K4]

ok. I did.
A permission denied error has occurred.

(tensorrt) nvidia@nvidia:/usr/src/tensorrt/samples/python/yolov3_onnx$ python onnx_to_tensorrt.py 
Downloading from https://github.com/pjreddie/darknet/raw/f86901f6177dfc6116360a13cc06ab680e0c86b0/data/dog.jpg, this may take a while...
Traceback (most recent call last):
  File "onnx_to_tensorrt.py", line 179, in <module>
    main()
  File "onnx_to_tensorrt.py", line 135, in main
    'https://github.com/pjreddie/darknet/raw/f86901f6177dfc6116360a13cc06ab680e0c86b0/data/dog.jpg', checksum_reference=None)
  File "/usr/src/tensorrt/samples/python/yolov3_onnx/yolov3_to_onnx.py", line 741, in download_file
    wget.download(link, local_path)
  File "/home/nvidia/virtualenv/tensorrt/local/lib/python2.7/site-packages/wget.py", line 506, in download
    (fd, tmpfile) = tempfile.mkstemp(".tmp", prefix=prefix, dir=".")
  File "/usr/lib/python2.7/tempfile.py", line 314, in mkstemp
    return _mkstemp_inner(dir, prefix, suffix, flags)
  File "/usr/lib/python2.7/tempfile.py", line 244, in _mkstemp_inner
    fd = _os.open(file, flags, 0600)
OSError: [Errno 13] Permission denied: './dog.jpgNnmJ7r.tmp'

[Hiroaki-K4]

By the way, do you use virtualenv for your environment?

[kumuji]

I think you should change permissions for the folder you are working in.
I prefer using virtual environment.

[Hiroaki-K4]

I already tried
chmod -R 777 trt_yolo_ros
But it doesn't success.

[kumuji]

I meant that for the /usr/src/tensorrt/samples/python/yolov3_onnx. Because if the example from nvidia wouldn't work for you, my code wouldn't work either. Can you check if examples work for you?

[Hiroaki-K4]

ok.
onnx_to_tensorrt.py resulted in an error on virtualenv. However, python3 onnx_to_tensorrt.py ran successfully locally.

I've probably figured out why it didn't work.
I tried to run the following command to create a symbolic link.

ln -s /usr/local/lib/python2.7/dist-packages/tensorrt $VIRTUAL_ENV/lib/site-packages/

However, there was no folder called tensorrt in /usr/local/lib/python2.7/dist-packages. So instead I ran the following command:

ln -s /usr/local/lib/python3.6/dist-packages/tensorrt $VIRTUAL_ENV/lib/site-packages/

I think this has an effect, what do you think? Is there any way to solve it?
I have one more question to ask. I want to run yolov4 accelerated by tensorrt on ros. Is it possible to combine trt_yolo_ros and tkDNN?
https://github.com/ceccocats/tkDNN

[kumuji]

Hmm, what Jetpack version are you using? I will check, maybe they dropped support for tensorrt in later vereions. Were you able to run the code linking that library? Also, are you sure that your $VIRTUAL_ENV is defined correctly(just in case)? As the result you should be able to un python -m pip show tensorrt and see the package there.
About tkDNN - should be possible, but you will have to write wrapper for that, it is not very compatible with code i have.

[Hiroaki-K4]

Jetpack version is 4.3. The command below could not be executed.

ln -s /usr/local/lib/python2.7/dist-packages/tensorrt $VIRTUAL_ENV/lib/site-packages/

When I run pip show tensorrt on vertualenv, the following message is displayed.

WARNING: Package(s) not found: tensorrt

[Hiroaki-K4]

Is it easier to make the trt_yolo_ros code yolov4 compatible than to combine tkDNN and trt_yolo_ros?
Also, do I need to edit the source code if I want to run trt_yolo_ros on python3?

[kumuji]

Jetpack version is 4.3. The command below could not be executed.

ln -s /usr/local/lib/python2.7/dist-packages/tensorrt $VIRTUAL_ENV/lib/site-packages/

When I run pip show tensorrt on vertualenv, the following message is displayed.

WARNING: Package(s) not found: tensorrt

You are right. I checked on jetpack 4.4 there is no package tensorrt anymore. Let me check where it is gone =)
Because in previous versions it was installed on python2.7 by default

[kumuji]

Is it easier to make the trt_yolo_ros code yolov4 compatible than to combine tkDNN and trt_yolo_ros?
Also, do I need to edit the source code if I want to run trt_yolo_ros on python3?

It is fully written in cpp, and my code is only python, so not very easy.
No, should be fine

[Hiroaki-K4]

ok!
Thank you for answering my question many times. Lastly, I would like to ask you one question. What would you do if you wanted to:
・ You want to run the faster yolov4 on ros
・Since there is already a source code for object detection that yolov3 moves on darknet_ros, You would like to implement it as close as possible.

Thanks!