Questions and answers
Subscribe to AI
Tricky Questions about the SURVEILLANCE REVOLUTION
1. Neural networks only work online, if the internet is down, everything will crash?
No, all neural networks are autonomous and operate on board the videoblazer. Disconnecting and connecting the network does not affect them in any way. And in general, videoblazer is about security. Any security issue is about reliability. Therefore, all the main functions work inside the unit.
2. This answer immediately brought up another question: You say that all the main functions work inside the unit, but is there something, even non-essential, that only works online - outside the unit?
Not at the moment, but it is planned. In particular, we will be connecting ChatGPT via the API provided by OpenAI. But these will only be service functions, such as voice control, convenient archive search, creation of analytical reports based on the smart archive, quick search for interesting moments and their relationships, etc.
3. What’s the revolution then?
The newest family of object detection models YOLOv8 doesn’t just work faster and more accurately than its older brothers, providing excellent recognition of even small objects and even in the dark - as its creators write about it laconically. The qualities that YOLOv8 gives us today, for us special labs who developed neural networks even under Tsar Pea, are mind-blowing. Never before have computers been so close to human-like vision of the world. Life does not stand still, surely in a couple of years we will be equally admiring something more grandiose, but at the moment YOLOv8 is already an opportunity for a machine to perform tasks at the same cognitive level as a human. That is, this is already enough for a computer decision to be taken as a human opinion. That’s it, now there’s no difference who makes the decision in a particular situation when it comes to seeing the world around us. Yes, mistakes are possible, but with the same probability as both a computer and a human.
Well, except that human intuition has not been reproduced. Most likely, trust in the machine will be undermined for a long time by this undefined term. What is intuition? A feeling of an informational nature, devoid of logic, is presented as the highest form of information accuracy. The computer will definitely not be able to handle this soon. Only Ivan the Fool can get “that, I don’t know what”. But… we’ll try, we live in Ivanovo after all, where there are many of them.
And, as in a fairy tale, we are accompanied by luck, YOLOv8 has settled very harmoniously on the newest chips, created specifically for neural network calculations. And now all this power is housed in a small box that replaces an expensive computer with a bunch of insanely expensive “video cards” for neural networks. You know perfectly well, even from our price list, how expensive video servers for neural networks are.
The main thing, of course, is not the budget, but for security you need as much reliability as possible and often the ability to use it in extreme conditions. You can’t hang huge video servers on poles, you can’t bury them in a damp shed, and videoblazer is a super-device for any type of object.
4. Judging by the price list, the price is not cheap?
The question is not posed correctly, not cheap in relation to what? For the cost of a flash drive it is expensive, for the nearest competitor - a powerful computer with three nVidia cards - it is 10 times cheaper. Let’s agree, it’s not a penny DVR. But the task is not to record how you are killed. Neural networks recognize the threat, and that, you agree, are incomparable concepts.
5. Your presentation has a lot of neural networks, and in the price list, most of them require additional thousands of rubles. Do you consider this cheap?
It seems you are talking about extras. For basic programs, basic neural networks are definitely integrated, which are needed by the vast majority of users, probably for 90% of the people. The rest are needed by large enterprises, which, moreover, need adaptation to their abnormal conditions. That’s why we set these prices for localization.
Naturally, when there are many such fine-tunings, they will be included in the standard delivery. Therefore, our prices should be understood in terms of development. Moreover, in the negative (not everything goes up).
6. We are developers in the field of YOLO. Can we somehow use your device for our own purposes, for our own neural networks?
We think it will be possible soon, we will prepare the API. You will simply buy the hardware - without paying for the software. Or you can write your own modifications to our basic software.
7. I liked the topic with the Journal of Operations. For security organizations, this presentation may be convenient. But it’s better to make all these cut-out figures of people and machines - with the arrangement of tiles across the entire screen, filling it as they arrive. Like in an image viewer. Don’t you think this is better?
Yes, the first version is made for security companies, this is the logic there. In the near future, there will be many varieties of output of the so-called “instant archive” (this name is still secret, so don’t tell anyone!).
But the most interesting part of this is that videoblazer, based on AI, will clean up the background. Then the objects will be immediately visible more clearly - without unnecessary details. Work is already underway, if you buy the device now - you will be able to upgrade for free.
8. The 8K in your advertising is cool, but I understand it’s just recording to a disk, which doesn’t really require any resources. What real resolution is output to the screen and analyzed? Like everyone else, do you take a second channel slightly below 1 MP, like D1?
It’s good to see specialists - connoisseurs of technology - contacting us. Until yesterday, all intelligent video surveillance systems worked exactly like that. Videoblazer is the first device that performs all actions with one channel - and this channel is of the maximum resolution. There is no possibility to enter a second or third channel in the videoblazer settings. It unpacks one video, analyzes it, writes it, and can repackage it for video messaging. It is also unpacked and output to monitors in 8K. Hardware chips allow this to be done without increasing the load on the main processor (unlike PCs).
By the way, you can use the second and third channels from the cameras for other systems and tasks.
But, naturally, there is a fly in the ointment. Our advertising talks about the total power - 8K, that is, it is for one channel. And for household systems, 8 channels of Full HD format or 12 channels of HD are used. But, agree, HD is 4 times larger than 1 MP. And almost 8 times larger than D1 - usually this resolution is taken for video analysis. Well, and when you look at the work, you are really amazed by the quality of recognition of the latest coolest neural networks against the background of full HD, and not a secondary channel of small size.
By the way, we already have orders for one 8K camera for a drone. Imagine what accurate recognition it gives from above! It is also convenient to assess the situation at a smart intersection with only one camera, so that everything is in one place.
9. Videoblazer is also a computer, isn’t it? What is the difference between a PC-based video surveillance system and yours?
Even a simple TV is also a computer. The architecture hasn’t changed since the creation of man or even animals: a computer, memory, power units, power supply system… :) But there are always differences.
Just as you can create many different systems on a personal computer, from an accounting server to a video surveillance system, you can also organize hundreds of cases of tasks on videoblazer. Only videoblazer differs from a PC in its hardware architecture - more suitable for video processing. There is hardware unpacking of each video channel, which favorably affects the quality of video streams. On a PC, all channels are processed by one CPU, and almost everything is done by the central processor, frantically switching between tasks, so the video constantly freezes, twitches, during the caching of operating systems everything freezes altogether. Videoblazer has independent chip-channels for unpacking, neural networks, analysis, sending video content stitched with metadata to client applications, as well as compressing this video, etc. All of the data received together is evaluated by the central processor of the Videoblazer, which in standard mode is not loaded more than 30% of its capacity. Therefore, under any peak load, Videoblazer will work normally on all tasks, without delays. And if you look at the CPU reading in a video surveillance system on a personal computer, it’s always over 100% 😊
New questions will be posted here…