Seemingly far from perfect, though I suppose a human has to go through it all and watch over any highlighted points of interest.
Some needs to watch termintor again… F
I have several questions.
Babushka’s gonna be pissed when from one second to the other she ends up in front of Saint Peter and he tells her that her house got bombed because some AI drone identified it as a heavy tank with 55% probability (last frame of the video).
Even if this is just a simulation using AI to classify a pre-recorded video sequence the fact is that this is the future. Western nations will want a human in the decision chain but I can think of many others who won’t give a toss.
Oh, and where combat footage?
It is impressive if implemented in drone itself, but it might be that it’s processed captured video from a drone: drones usually don’t transmit digital HD video, but rather record it on a flash memory, and here we can see clear digital video.
“The machines rose from the ashes of the nuclear fire.”
i saw this in an arma 3 mod, behaviour was exactly the same with the AI misidentifying some bush for a tank
This is pretty basic hobby level detection, not that interesting yet, but we will see where this evolves to
This is just a showcase of basic image recognition. It might not even be footage from an actual drone equipped with this technology, just a video run through the software. It looks like a super early proof of concept rather than something useful, as it doesn’t even recognize the “light armor” without zooming in, and even then with only low 20-percentile confidence. Not good.
It’s also not combat footage.
Next level!
This is likely using something like YOLO or similar object detection algorithm. For anyone wondering what the numbers are above each detection box, it is usually the confidence score of the detection.
I believe if they scaled the objects (tanks) a little smaller during the training process, they may be able to get some better, more reliable detections at the zoomed out FOV.
Awesome glimpse at what is being used and I would bet that is a home grown solution to solve their problem. Pretty cool.
we already have seen both sides already using this kind of target tracking.
I can make this for you in about 10 minutes.
Let me know if you are interested.
$1 million flat rate.
14 comments
Seemingly far from perfect, though I suppose a human has to go through it all and watch over any highlighted points of interest.
Some needs to watch termintor again… F
I have several questions.
Babushka’s gonna be pissed when from one second to the other she ends up in front of Saint Peter and he tells her that her house got bombed because some AI drone identified it as a heavy tank with 55% probability (last frame of the video).
Even if this is just a simulation using AI to classify a pre-recorded video sequence the fact is that this is the future. Western nations will want a human in the decision chain but I can think of many others who won’t give a toss.
Oh, and where combat footage?
It is impressive if implemented in drone itself, but it might be that it’s processed captured video from a drone: drones usually don’t transmit digital HD video, but rather record it on a flash memory, and here we can see clear digital video.
“The machines rose from the ashes of the nuclear fire.”
i saw this in an arma 3 mod, behaviour was exactly the same with the AI misidentifying some bush for a tank
This is pretty basic hobby level detection, not that interesting yet, but we will see where this evolves to
This is just a showcase of basic image recognition. It might not even be footage from an actual drone equipped with this technology, just a video run through the software. It looks like a super early proof of concept rather than something useful, as it doesn’t even recognize the “light armor” without zooming in, and even then with only low 20-percentile confidence. Not good.
It’s also not combat footage.
Next level!
This is likely using something like YOLO or similar object detection algorithm. For anyone wondering what the numbers are above each detection box, it is usually the confidence score of the detection.
I believe if they scaled the objects (tanks) a little smaller during the training process, they may be able to get some better, more reliable detections at the zoomed out FOV.
Awesome glimpse at what is being used and I would bet that is a home grown solution to solve their problem. Pretty cool.
we already have seen both sides already using this kind of target tracking.
I can make this for you in about 10 minutes.
Let me know if you are interested.
$1 million flat rate.