opendr
opendr copied to clipboard
opendr-eu
Tx2 install
TX-2 installation script. Updated installation.md and fixed some import statements in object_detection_2d/detr and object_detection_2d/gem
I don't have a TX2 to test this, can somebody else validate that the script works? @thomaspeyrucain maybe?
I don't have a TX2 to test this, can somebody else validate that the script works? @thomaspeyrucain maybe?
@ad-daniel Ok I will test it on a docker on our Jetson
Just to be sure I wanted to check whether this dockerfile was affected by the nvidia key rotation but it seems that it isn't possible to build the image without arm system (or at least, I wasn't able to despite passing --platform linux/arm64 when building), so I haven't been able to test. Has anybody managed to build successfully the image the past 2-4 days? Or did you encounter a key issue?
Just to be sure I wanted to check whether this dockerfile was affected by the nvidia key rotation but it seems that it isn't possible to build the image without arm system (or at least, I wasn't able to despite passing
--platform linux/arm64when building), so I haven't been able to test. Has anybody managed to build successfully the image the past 2-4 days? Or did you encounter a key issue?
Since the last build on our side happened last week, I am running the docker build again now. Will update when it is completed.
Just to be sure I wanted to check whether this dockerfile was affected by the nvidia key rotation but it seems that it isn't possible to build the image without arm system (or at least, I wasn't able to despite passing
--platform linux/arm64when building), so I haven't been able to test. Has anybody managed to build successfully the image the past 2-4 days? Or did you encounter a key issue?
Just built the docker image and it is working.
@Pavlos-Tosidis Just a last question, did you build the images on your machine or the TX2? Because if like me you are unable to build it on a amd64 platform, then it means we won't be able to have the CI create the image and publish it automatically (as we do for the other docker images)
It showed some errors during the process but it finished building the image OpenDR_docker_build_NX.txt
I tried to install packages for running the ROS nodes : sudo apt-get install ros-noetic-vision-msgs ros-noetic-geometry-msgs ros-noetic-sensor-msgs ros-noetic-audio-common-msgs But it was not able to find the packages so I cloned the repositories and then built the ws
Looks like some packages were not well installed or is missing a library :
root@98fb18d18058:/opendr# python3
Python 3.6.9 (default, Mar 15 2022, 13:55:28)
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import mxnet
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/opendr/mxnet/python/mxnet/__init__.py", line 23, in <module>
from .context import Context, current_context, cpu, gpu, cpu_pinned
File "/opendr/mxnet/python/mxnet/context.py", line 23, in <module>
from .base import classproperty, with_metaclass, _MXClassPropertyMetaClass
File "/opendr/mxnet/python/mxnet/base.py", line 351, in <module>
_LIB = _load_lib()
File "/opendr/mxnet/python/mxnet/base.py", line 341, in _load_lib
lib_path = libinfo.find_lib_path()
File "/opendr/mxnet/python/mxnet/libinfo.py", line 73, in find_lib_path
'List of candidates:\n' + str('\n'.join(dll_path)))
RuntimeError: Cannot find the MXNet library.
List of candidates:
/opendr/lib/libmxnet.so
/opt/ros/noetic/lib/libmxnet.so
/usr/local/cuda-10.2/targets/aarch64-linux/lib/libmxnet.so
libmxnet.so
/opendr/mxnet/python/mxnet/libmxnet.so
/opendr/mxnet/python/mxnet/../../lib/libmxnet.so
/opendr/mxnet/python/mxnet/../../build/libmxnet.so
../../../libmxnet.so
>>> import torch
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/local/lib/python3.6/dist-packages/torch/__init__.py", line 189, in <module>
_load_global_deps()
File "/usr/local/lib/python3.6/dist-packages/torch/__init__.py", line 142, in _load_global_deps
ctypes.CDLL(lib_path, mode=ctypes.RTLD_GLOBAL)
File "/usr/lib/python3.6/ctypes/__init__.py", line 348, in __init__
self._handle = _dlopen(self._name, mode)
OSError: libcurand.so.10: cannot open shared object file: No such file or directory
Looks like some packages were not well installed or is missing a library :
root@98fb18d18058:/opendr# python3 Python 3.6.9 (default, Mar 15 2022, 13:55:28) [GCC 8.4.0] on linux Type "help", "copyright", "credits" or "license" for more information. >>> import mxnet Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/opendr/mxnet/python/mxnet/__init__.py", line 23, in <module> from .context import Context, current_context, cpu, gpu, cpu_pinned File "/opendr/mxnet/python/mxnet/context.py", line 23, in <module> from .base import classproperty, with_metaclass, _MXClassPropertyMetaClass File "/opendr/mxnet/python/mxnet/base.py", line 351, in <module> _LIB = _load_lib() File "/opendr/mxnet/python/mxnet/base.py", line 341, in _load_lib lib_path = libinfo.find_lib_path() File "/opendr/mxnet/python/mxnet/libinfo.py", line 73, in find_lib_path 'List of candidates:\n' + str('\n'.join(dll_path))) RuntimeError: Cannot find the MXNet library. List of candidates: /opendr/lib/libmxnet.so /opt/ros/noetic/lib/libmxnet.so /usr/local/cuda-10.2/targets/aarch64-linux/lib/libmxnet.so libmxnet.so /opendr/mxnet/python/mxnet/libmxnet.so /opendr/mxnet/python/mxnet/../../lib/libmxnet.so /opendr/mxnet/python/mxnet/../../build/libmxnet.so ../../../libmxnet.so >>> import torch Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/usr/local/lib/python3.6/dist-packages/torch/__init__.py", line 189, in <module> _load_global_deps() File "/usr/local/lib/python3.6/dist-packages/torch/__init__.py", line 142, in _load_global_deps ctypes.CDLL(lib_path, mode=ctypes.RTLD_GLOBAL) File "/usr/lib/python3.6/ctypes/__init__.py", line 348, in __init__ self._handle = _dlopen(self._name, mode) OSError: libcurand.so.10: cannot open shared object file: No such file or directory
Dear Thomas, Did you follow the instructions for building the docker image correctly? This happened to me when I didn't edit the '/etc/docker/daemon.json' file correctly. Additionaly the script builds the ros packages as well, so there is no need to install further packages.
@Pavlos-Tosidis I discussed this with my colleague and we checked the log, it is missing some things inside the docker like cudnn.h and libcurand.so.10 could you please check that everything is correctly installed in your docker?
@Pavlos-Tosidis I discussed this with my colleague and we checked the log, it is missing some things inside the docker like cudnn.h and libcurand.so.10 could you please check that everything is correctly installed in your docker?
@thomaspeyrucain as we discussed, I have flashed and installed the docker image multiple times on a TX2, with the results being the same. I didn't try the ROS scripts, but the python scripts run successfully.
Do you build the docker image from scratch ? This base image : nvcr.io/nvidia/l4t-base:r32.6.1 Does not contain the cudnn.h and the corresponding libraries
Do you build the docker image from scratch ? This base image : nvcr.io/nvidia/l4t-base:r32.6.1 Does not contain the cudnn.h and the corresponding libraries
Do you run the container with the proper arguments? -it --privileged
The errors are occurring when building the docker from the Dockerfile Those arguments -it --privileged is when you run the docker image
The errors you sent me so far are while trying to import torch/mxnet. There are a lot of warnings when building mxnet, but it gets installed and working as intended.
It showed some errors during the process but it finished building the image OpenDR_docker_build_NX.txt
If you check the logfile errors are appearing when building the docker image
It showed some errors during the process but it finished building the image OpenDR_docker_build_NX.txt
If you check the logfile errors are appearing when building the docker image
Since it produces the docker image in the end. Can you please try and run for example (after activating opendr): python3 projects/perception/face_recognition/demos/infrence_demo.py
@thomaspeyrucain Just finished the docker image installation on an Nvidia NX. Everything run smoothly and tried face_recognition and Retinaface inference demos for testing. Both run as intended.
@Pavlos-Tosidis Nice, Did you succeed to run the ROS nodes without issues inside the docker?
NX docker image is up on docker hub. you can download and run it with: xhost +local:root sudo docker run -it --privileged -v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=unix$DISPLAY opendr/opendr-toolkit:nx /bin/bash
Fixed a couple of ROS issues. If the NX is correctly flashed (with Cuda and nvidia runtime) it works out of the box
I pulled the image and tested it on the NX board and it works for the face recognition demo script What ROS issues did you fix?
The broken tests on macos can be ignored, cppcheck was upgraded and it's being fixed on a separate PR
