Installing opencv 3.0.0 on the Jetson Tk1

Autonomous Machines Jetson & Embedded Systems Jetson TK1
#1

Like the rest of what was installed using jetpack, I assume it will have to be cross-compiled?
I’ve installed opencv3.0.0 before on my debian laptop, but I’m not sure of myself in that I can install opencv3.0.0 on my Jetson without something going wrong.

Can someone that has successfully installed the opencv3.0.0 beta explain to me how you did it on the jetson board?

#2

Yes, I’ve done it.
Follow the instructions on this page: http://elinux.org/Jetson/Installing_OpenCV in the section “Natively compiling the OpenCV library from source onboard the device”.
Use this download link: https://github.com/Itseez/opencv/archive/3.0.0-beta.zip instead of the one on that page (that one’s for 2.4.9).
You will have to replace the ‘2.4.9’ version numbers in the installation commands to match your files (the 3.0.0-beta zip you downloaded fro above.)

#3

"Note: Compiling OpenCV from source will not give you NVIDIA’s CPU optimizations that are only available in the closed-source prebuilt OpenCV4Tegra packages. "

How significant is this difference in speed?

#4

I used both “standard OpenCV” (from Ubuntu repo) and “opencv4tegra” and I noticed 4-5 fps of difference running the same algorithm (not using GPU).
I believe that opencv4tegra has a lot of optimization related to Tegra architecture

#5

I have built a opencv-3.1.0 version that runs fine on my TK1, which performs better than Tegra-2.4.12.3 for most of my application, but some parts remain faster with 2.4.12.3, so I’d like to test with Tegra optimizations.

Does anyone knows if a release of closed-source Tegra OpenCv 3.x.x is planned or not ?
Or a plan to open this source ? (is there more than round.hpp and build options ?)

#6

Hi Honey_Patouceul,

To optimize OpenCv 3.x.x for Tegra is not in current plan yet.
May I know the which part is worse on OpenCV4Tegra 2.4.12.3?
And which part has better performance?

Thanks

#7

Just for instance, the BRISK keypoint detector and descriptor extractor are more robust in 3.1.0

#8

I just installed OpenCV 3.0.0 using the instructions above.
It appears to have installed successfully, but I get the following error:

ubuntu@tegra-ubuntu:~$ python
Python 2.7.6 (default, Mar 22 2014, 23:30:12)
[GCC 4.8.2] on linux2
Type “help”, “copyright”, “credits” or “license” for more information.

import cv2
Traceback (most recent call last):
File “”, line 1, in
ImportError: No module named cv2

Any ideas?

#9

Hello, x92127:
Please take a look at http://stackoverflow.com/questions/19876079/opencv-cannot-find-module-cv2.
it may help.

br
ChenJian

#10

I solve the problem by using different install instructions, found here:
http://docs.opencv.org/3.0-beta/doc/tutorials/introduction/linux_install/linux_install.html

#11

Hi,
I’m new to this Forum and installed latest Jetpack with libs on TK1, as opencv2.

Anyway, all projects, books and sources are based on openCV3.

  • Is the answer from kayccc ( openCV3 for tegra not available) still true?
  • if yes : if I use the solution from cn04 (installing & compile from source) aside the current opencv4tegra (opencv2) installation = could they live together or the last installation from source would overwrite it?
#12

Please see the reply at below thread:

https://devtalk.nvidia.com/default/topic/976565/jetson-tk1/having-opencv3-with-cuda-onjetson-tk1/post/5020998/#5020998

#13

Does a plan to optimize opencv 3.x exist now?
Or we will be stuck on the 2.x version?

Cheers
Bruno

#14

There is still no plan to migrate on OpenCV3.x.

Thanks