Awesome video Ray. I'm happy to see that it seems to be stable and has been running for a few days.
One quick comment about frame rates though.
So, the frame rate that you're seeing from OpenCV's frame grabber is likely 15 or maybe 30FPS.. The rate that Yolo is actually recognizing frames might be closer to 3 to 4 frames per second. The way it's implemented, Yolo will skip frames if it's not done processing the current one. Once it's done, it adds it's boxes to the current frame that is being displayed.
The nice advantage of this approach is that the normal frame rate for OpenCV isn't affected (much) when Yolo is or isn't running.
The disadvantage is that the boxes that are drawn might be a few frames out of synch with their actual location.
Going to be fun to see how this runs on my Nvidia jetson tx2,
Have just finnished porting / re writing drivers for my dual rassberry pi V 2.1 8mp cam's,
There where no drivers for them in the more resent version of the kernel / Jetpack , time consuming stuff :/
Also batteling problems getting Myrobotlab to se my ttyACMx ports, i am using arduino due's and they work just fine in windows, Myrobotlab runs good on the TX2, only problem is getting it to se the ttyACMx :(
Great work to all off you innvolved in this :)
Btw, i am installing opencv 3.4.0 and YOLO3 (DarkNet) , will that be ok with the new / next version og Myrobotlab ?
When it comes to Opencv, would opencv 3.4.0 and YOLO3 (DarkNet) be ok with MyRobotLab, current and the upcoming Nixie release, or does it need a spesial version ?
I am pretty new at MyRobotLab, so havent go a feell yet how this is stitched togehther :)
And last quiestion, would Tensorflow / RT work with MyRobotLab ?
So.. the OpenCV service in MyRobotLab (on the nixie release) is JavaCV version 1.4.0 (soon to be 1.4.1) which translates into including OpenCV 3.4.0 (soon to be OpenCV 3.4.1 )
It does have the ability to load machine learning models from various open source platforms. (Caffe, Tensorflow, Darknet, and Torch currently from what I can tell.)
There is no need to install opencv, it's included with MyRobotLab in the OpenCV service.
We currently have an opencv filter for YoloV2 (as this is what was easy to include at the time.) I guess there's not much difference in accuracy between YoloV2 and YoloV3, but I guess the performance is better with V3.
Wonderfull, Very very nice.
Wonderfull, Very very nice.
frame rate
Awesome video Ray. I'm happy to see that it seems to be stable and has been running for a few days.
One quick comment about frame rates though.
So, the frame rate that you're seeing from OpenCV's frame grabber is likely 15 or maybe 30FPS.. The rate that Yolo is actually recognizing frames might be closer to 3 to 4 frames per second. The way it's implemented, Yolo will skip frames if it's not done processing the current one. Once it's done, it adds it's boxes to the current frame that is being displayed.
The nice advantage of this approach is that the normal frame rate for OpenCV isn't affected (much) when Yolo is or isn't running.
The disadvantage is that the boxes that are drawn might be a few frames out of synch with their actual location.
Run Time
Hello kwatter,
Even a low processed frame rate with the yolo is still good.
The best part is the duration of the running of the yolo. Now wer are looking at in-excess of 3 days without a break.
This proves the build is stable, and that is almost as important as the rate that it runs at.
Ray
Just an update
The test version of Nixie shown in the video has now been running a full week non stop running Yolo.
The service is still running and I'm still around the 80% mark a Person :-)
It has processed 9,063,000 frames during its run time.
I don't care what you say, i think that is impressively stable.
Very Nice :)
Going to be fun to see how this runs on my Nvidia jetson tx2,
Have just finnished porting / re writing drivers for my dual rassberry pi V 2.1 8mp cam's,
There where no drivers for them in the more resent version of the kernel / Jetpack , time consuming stuff :/
Also batteling problems getting Myrobotlab to se my ttyACMx ports, i am using arduino due's and they work just fine in windows, Myrobotlab runs good on the TX2, only problem is getting it to se the ttyACMx :(
Great work to all off you innvolved in this :)
Btw, i am installing opencv 3.4.0 and YOLO3 (DarkNet) , will that be ok with the new / next version og Myrobotlab ?
Ahoy Spawn32 ! Regarding the
Ahoy Spawn32 !
Regarding the ttyACMx ports.
$ ls /dev/tttyACM* comes back with stuff ?
root would...
Thanks for the tip
Thanks for the tip GrogG,
Both ports show up in /dev/ but it might like you say be a user access thing, will check it up tonight.
Right now i am puting a Articulating Neck on Ada :)
When it comes to Opencv,
When it comes to Opencv, would opencv 3.4.0 and YOLO3 (DarkNet) be ok with MyRobotLab, current and the upcoming Nixie release, or does it need a spesial version ?
I am pretty new at MyRobotLab, so havent go a feell yet how this is stitched togehther :)
And last quiestion, would Tensorflow / RT work with MyRobotLab ?
javacv includes opencv
Hi Spawn32
So.. the OpenCV service in MyRobotLab (on the nixie release) is JavaCV version 1.4.0 (soon to be 1.4.1) which translates into including OpenCV 3.4.0 (soon to be OpenCV 3.4.1 )
It does have the ability to load machine learning models from various open source platforms. (Caffe, Tensorflow, Darknet, and Torch currently from what I can tell.)
There is no need to install opencv, it's included with MyRobotLab in the OpenCV service.
We currently have an opencv filter for YoloV2 (as this is what was easy to include at the time.) I guess there's not much difference in accuracy between YoloV2 and YoloV3, but I guess the performance is better with V3.
Nice :) I would probaly need
Nice :)
I would probaly need a recompiled opencv service with the Cuda / Cuda9=1 flag set then :)
Will test this out as soon as i can connect with my Due's in MyRobotLab...