5 POINTS about OPENCV on Jetson Dev Kits

hello it's Jim from Jessa nice calm on today's show I want to share with you five different things about OpenCV on the Nvidia Jets and developer kits we get a lot of questions about OpenCV on the Justin heck's channel let me share five important points with you the first one is how do we figure out what build is running and what is available we typically see this type of question has does OpenCV run CUDA at this OpenCV run gstreamer does OpenCV look like a cat we don't get that one as much mmm we've never gotten that one let's figure out what version we are running of OpenCV this is a Jetson nano here's our environment we have just flashed a brand new Jetson nano image we are running L 40 30 2.2.1 from a jetpack 4.2.2 the numbers are funny the usual way that we look up the package version is to use d package let's do that - l and let's wrap live open CV we can see that we're running wipe open CV version 3.3 point 1 I would assume this is a commit version that gives us what version we are running open CV is a pretty complex environment we'd also like to know some more about the build information things like is GStreamer supported is kuta supported what modules have been built and so on we can import the opencv library into python and look at the build information let's do that python import cv - that is the name of the opencv library and now we are going to print the build information and it basically dumps how it built OpenCV we can see that it's OpenCV three point three point one you see the platform so we can see it was a linux four point nine Tegra the see make version release version we can see the CPU hardware features neon for example on the Jetson that's a default on the 64-bit CPUs we can see all the compiler flags for C and C++ in which compiler they used here are the open CP modules the gooey goo gtk+ there's no QT supports built in here's some information about the media IO video IO we can see that GStreamer is built in version one point 14.1 we can see here and the other third-party libraries which one we used we use Diegan and here's one of the most frequently asked questions use cuda the answer is no and we can see here that we have python 2 and python 3 support now we can also see that there are tests and samples available that were built and the install path is slash user let's open up another terminal let's wander over to user we can see some of the binaries here OpenCV version let's go over to the library we can see that there are a bunch of OpenCV libraries here and let's get out of Python the second question that we can asked a lot is does this version of OpenCV work with Python three the answer is yes let's run Python 3 let's import the library cb2 and we get an error hmm well that's not good if you are familiar with Python you would probably guess that numpy is not installed we see that there's an importer numpy and it failed to import let's take a look at our modules this will list all of the installed modules ok wander up here a little bit we can see that macaron bakery is installed that's important but no numpy I don't see it let's install it open up the terminal and we need the Python 3 version if you're more comfortable using pip 3 you could use that also that looks like it went ok let's close that terminal let's give her another whirl import CV - oh it looks happy now and let's take a look at our build information sure enough so we're up and running in Python 3 the third question usually is does this version of OpenCV have a feature in particular GStreamer and CUDA we look at our build information and we can answer that we look at video i/o gstreamer and we can see that version one point 14.1 is supported and in other third-party libraries we see that use CUDA is no this version of OpenCV the default version that ships with the Jetsons does not have CUDA support basically what that means is that if you want to CUDA support then you will need to compile OpenCV from source the fourth point is that if you are using open CV Python and CUDA you will need to use open CV version 4 version 3 of open CV does not support Python wrappers now if you're sufficiently motivated you can probably get to work with open CV 3 somehow but most people just go to open CV 4 five a lot of other applications don't play well with live open CV installations many applications Ross is a perfect example of this use the version of open CD that's in the abouttwo repository and that's I believe that's version 3.2 one of the issues there is that version 3.2 in the repository does not have built-in gstreamer support so that means that your TSI camera won't work or whatever a lot of people experience this and they don't know what happened and just sighs it doesn't work and like everybody else what happens is they start a long and arduous journey in installing and uninstalling hoping and praying that it will eventually work we've all been down that path check your version of open CV to make sure you know what's going on and here's the bonus point one of the things that people report after installing the gpu-accelerated OpenCV functions is that their algorithms do not run any faster this can be for a variety of reasons the first and foremost is that all OpenCV functions are not GPU accelerated so that's one thing to check the second reason is that some algorithms do not lend themselves to parallelization I think that's how you say it in other words it doesn't help to run the algorithm in parallel the third reason is that you're processing pipeline may not take advantage of parallel processing and the fourth reason is that bandwidth overhead may be a problem so for example it may take as long to get the data to the GPU and back as it does to get the acceleration benefit let's sum up those points the benefit that you get from moving your processing to the GPU on open CV will depend on the functions that you call the configuration that you use and the processing pipeline if you liked this video give it a thumbs up and if you have not already please subscribe thanks for watching

Channel: JetsonHacks

Views: 9,810

Keywords: JetsonHacks, Jetson AGX Xaver, Jetson Nano, Jetson TX1, Jetson TX2, OpenCV, OpenCV Python, Jetson Developer Kit, Jetson Development Kit

Id: dey9uTXuTo8

Length: 10min 2sec (602 seconds)

Published: Fri Nov 15 2019