They use one WiFi AP as a source of RF "light", and another as a wery poor quality multichannel SDR. That is why they had to put transmitting router and receiving router directly one against other and then use camera to train ANN to "recognize" several distinct combinations of people poses from this poor SDR in this exact configuration.
So to surveil over Soros NGO rithuals you have hack at least two WiFi routers (and one of routers should be in client mode, so will not work as AP and it will be noticed immidiately), put camera in a room, and force satanists to roam and take different poses in the room to train ANN by comparing image from camera and signals from "receiver" WiFi router for a few hours at minimum. Only then you will be able to "see" what is going in the room using only this two WiFi routers without using camera. And with huge uncertainity, because it is impossible with such poor multichannel SDR ("receiver" WiFi router) to properly cover all possible poses and positions of different number of people. Too little data from CSI, so you will get same signal patterns for multiple combinations.
It is like using 3x3 large pixel (three by three large pixels, total 9 pixels with size of, say, feet by feet) two-color (say, only R & G from usual RGB triplet) camera and try to figure out what it see using ANN trained comparing that 9 pixels data with high-resolution image from real camera. They even post this picture with 3x3 tensor wich represents what data they working with.
You could train ANN to produce some believable results sometimes, but it is still just not enough data to declare it as something useable.
And it will completely fail if someone will slightly move one of WiFi routers or even just because something changed in RF environment around, like another WiFi router added in a building or even somebody turn on microwave owen in dinner room.
And main thing - if you could do all of above including putting a high resolution camera in a Soros NGO room, then you obviously don't need to do all that WiFi "magic" at all, since you already will have a camera under your control in the room. :)
You could do that without ANN and so installing camera inside surveilled room, and with only one WiFi router inside, but you will have to use proper multichannel SDR receiver nearby that will give you much more data to analyse and create real, calculated, not "AI" generated picture of room internals.
Again, it is just another "AI" hype, nothing more. "Look, we could train ANN to produce believable pictures from 3x3 pixel sensor!"
If above is too complicated to get a clue, try something like "AI" that draw a picture from few words of description. It could produce something reasonable for descriptions and images it was trained for, but will fail to do anything out of script. Too little data on input to make it really working.
Did you read the paper?
They use one WiFi AP as a source of RF "light", and another as a wery poor quality multichannel SDR. That is why they had to put transmitting router and receiving router directly one against other and then use camera to train ANN to "recognize" several distinct combinations of people poses from this poor SDR in this exact configuration.
So to surveil over Soros NGO rithuals you have hack at least two WiFi routers (and one of routers should be in client mode, so will not work as AP and it will be noticed immidiately), put camera in a room, and force satanists to roam and take different poses in the room to train ANN by comparing image from camera and signals from "receiver" WiFi router for a few hours at minimum. Only then you will be able to "see" what is going in the room using only this two WiFi routers without using camera. And with huge uncertainity, because it is impossible with such poor multichannel SDR ("receiver" WiFi router) to properly cover all possible poses and positions of different number of people. Too little data from CSI, so you will get same signal patterns for multiple combinations.
It is like using 3x3 large pixel (three by three large pixels, total 9 pixels with size of, say, feet by feet) two-color (say, only R & G from usual RGB triplet) camera and try to figure out what it see using ANN trained comparing that 9 pixels data with high-resolution image from real camera. They even post this picture with 3x3 tensor wich represents what data they working with.
You could train ANN to produce some believable results sometimes, but it is still just not enough data to declare it as something useable.
And it will completely fail if someone will slightly move one of WiFi routers or even just because something changed in RF environment around, like another WiFi router added in a building or even somebody turn on microwave owen in dinner room.
And main thing - if you could do all of above including putting a high resolution camera in a Soros NGO room, then you obviously don't need to do all that WiFi "magic" at all, since you already will have a camera under your control in the room. :)
You could do that without ANN and so installing camera inside surveilled room, and with only one WiFi router inside, but you will have to use proper multichannel SDR receiver nearby that will give you much more data to analyse and create real, calculated, not "AI" generated picture of room internals.
Again, it is just another "AI" hype, nothing more. "Look, we could train ANN to produce believable pictures from 3x3 pixel sensor!"
If above is too complicated to get a clue, try something like "AI" that draw a picture from few words of description. It could produce something reasonable for descriptions and images it was trained for, but will fail to do anything out of script. Too little data on input to make it really working.