D415 with high CPU usage on Jetson NANO

[103061634]

• Before opening a new issue, we wanted to provide you with some useful suggestions (Click "Preview" above for a better view):

  • Consider checking out SDK examples.
  • Have you looked in our documentations?
  • Is you question a frequently asked one?
  • Try searching our GitHub Issues (open and closed) for a similar issue.

• All users are welcomed to report bugs, ask questions, suggest or request enhancements and generally feel free to open new issue, even if they haven't followed any of the suggestions above :)

---

Required Info
Camera Model	{ D400 }
Firmware Version	05.12.14.50
Operating System & Version	18.04
Kernel Version (Linux Only)	4.9.140-tegra
Platform	NVIDIA Jetson NANO
SDK Version	2.
Language	python
Segment	others

Issue Description

We observed high CPU usage when we just use python to capture depth(640x480 30fps) and color frame(640x480 30fps). But I think realsense should not compute something on CPU.
What is the cause of this, or any settings need to adjust. Any suggestions?
Thank you for your help.

Here is our cmake command:

cmake .. -DBUILD_EXAMPLES=true
-DCMAKE_BUILD_TYPE=release
-DFORCE_RSUSB_BACKEND=false
-DBUILD_WITH_CUDA=true
-DBUILD_PYTHON_BINDINGS:bool=true
-DPYTHON_EXECUTABLE=/usr/bin/python3
-BUILD_WITH_OPENMP=true
&& make -j$(($(nproc)-1)) && sudo make install

Here is our python script:

import pyrealsense2 as rs
import numpy as np
import cv2

pipeline = rs.pipeline()
config = rs.config()
config.enable_stream(rs.stream.depth, 640, 480, rs.format.z16, 30)
config.enable_stream(rs.stream.color, 640, 480, rs.format.bgr8, 30)


pipe_profile = pipeline.start(config)

try:
    while True:
        frames = pipeline.wait_for_frames()
        depth_frame = frames.get_depth_frame()
        color_frame = frames.get_color_frame()
        if not depth_frame or not color_frame:
            continue
        depth_image = np.asanyarray(depth_frame.get_data())
        color_image = np.asanyarray(color_frame.get_data())
finally:
    pipeline.stop()

[MartyG-RealSense]

Setting BUILD_WITH_OPENMP to True to take advantage of multiple cores when utilizing the librealsense functions listed in the image below can result in greater CPU usage, as described in the documentation for CMake build flags in this link:

https://dev.intelrealsense.com/docs/build-configuration

[103061634]

Thx for your response, we already make with -BUILD_WITH_OPENMP=true.
Is this( CPU usage) look like something wrongs?

[MartyG-RealSense]

What the description about BUILD_WITH_OPENMP means is that if you have it set to True then that is what can cause high CPU usage. So it would be an expected behaviour when including BUILD_WITH_OPENMP=true in your CMake build instruction.

[103061634]

We appreciate you for your help, it is work!!!

Is there have another reduce CPU usage measure?

[MartyG-RealSense]

The RealSense SDK supports GLSL, an alternative method of offloading work from the CPU to the GPU. It is 'vendor neutral', meaning that it should work with any brand of video chip. The improvement may not be noticable on low-power devices though, and there is not much information about using it with Python instead of C++.

#3654
https://github.com/IntelRealSense/librealsense/tree/master/examples/gl

Advice is given in a Jetson Nano case in the link below that using GLSL may not be faster than CUDA anyway.

#7824 (comment)

So using -DBUILD_WITH_CUDA=true and not setting -BUILD_WITH_OPENMP to true may provide the best results.

The discussion linked to above does provide a comment from the RealSense user in that case that they were able to achieve a 10x speed increase in their code using a method described in the link below.

#7824 (comment)

[Try-Hello]

I have a question, can you help me?

I use pyrealsense2-aarch64 in jetson nano. I use D455。

but：
import pyrealsense2 as rs
#And then define 'pipeline':
pipeline = rs.pipeline()

error：Module 'pyrealsense2' has no attribute 'pipeline'

how to solve this problem？please。

[MartyG-RealSense]

Hi @Try-Hello I will respond to your question at the new case that you have created for it. Thanks!

[103061634]

The RealSense SDK supports GLSL, an alternative method of offloading work from the CPU to the GPU. It is 'vendor neutral', meaning that it should work with any brand of video chip. The improvement may not be noticable on low-power devices though, and there is not much information about using it with Python instead of C++.

#3654
https://github.com/IntelRealSense/librealsense/tree/master/examples/gl

Advice is given in a Jetson Nano case in the link below that using GLSL may not be faster than CUDA anyway.

#7824 (comment)

So using -DBUILD_WITH_CUDA=true and not setting -BUILD_WITH_OPENMP to true may provide the best results.

The discussion linked to above does provide a comment from the RealSense user in that case that they were able to achieve a 10x speed increase in their code using a method described in the link below.

#7824 (comment)

Thx for your help!!! We will do it.