Developers of smart cameras, smart DVRs, and neural-network video analytics for surveillance systems need AI models capable of operating in real-world street conditions. Out there, nobody walks around with professional cameras, carefully adjusts angles, sets up lighting, records without compression, or follows the common sense taught in cinematography textbooks.
Of course, Gambit can be used for many other tasks, but its main focus is the convenient collection of material FROM video surveillance systems and dataset annotation specifically FOR video surveillance neural networks.
Gambit is not designed for polished photos and Internet reels. Quite the opposite — it is intended for low-quality surveillance archive footage. At SpesLab, we call this kind of content “wild.”
Searching for and preparing training material often takes more time than working with it itself. One of Gambit’s most useful features is that it allows you to load archives directly on the fly:
You can specify a folder containing video recordings, and Gambit will automatically scan the files and extract frames at a chosen interval.
2. You can also select an individual video file.
3. Or you can immediately begin annotating a folder of images.
4. Another useful addition is content input through a USB video camera. From time to time, urgent requests arrive for neural networks capable of recognizing some completely new object provided physically by a customer — perhaps an aircraft component or a badge.
Without any complicated filming or video editing workflows, you simply turn on the laptop camera and start. Gambit immediately begins preparing the material.
Yes, sometimes this produces a lot of unnecessary data, but it can be filtered quickly during annotation. Besides, at the early stages nobody truly knows what is unnecessary. Frames without objects are also extremely important, because without background data the neural network may behave incorrectly.
In fact, after many experiments and many mistakes, we concluded that for surveillance neural networks, isolated cut-out objects work poorly. Everything must be annotated together with the background. I know many people disagree with this, but we follow our own convictions.
And as a “doctor,” I would also like to warn you: the quality of a neural network depends not on the annotation software, but on the methodology.
SpesLab has developed — or perhaps suffered through — many methodologies that we could talk about. Although I have noticed a strange feature on this platform: everyone wants to learn everything, but at the same time creates conditions where publishing becomes almost impossible. Quite an original fetish.
And quality also depends on careful hands. Add to that a sharp eye and fine motor skills — qualities often associated with women. That is why SpesLab designed this software for housewives. (This is actually a reference to Lenin’s work, so gentlemen, please don’t be offended!)
Gambit includes smooth border adjustment in several modes, both with a mouse and with keyboard arrows. When working with it, you feel like an artist. Or even like a child, because the process resembles coloring pictures. It is genuinely mesmerizing.
It is also surprisingly calming, so I recommend it as an evening activity after work.
Please forgive my feminine perspective on technology. Things are a little too serious around here — articles rarely contain any emotion. I thought I would at least add some life to it.
I am still studying at university myself, but older women on our team — especially mothers — experience almost maternal feelings when the second stage of annotation begins: fine-tuning.
That is when the trained neural network already highlights objects by itself, and you only need to correct the little mistakes. As they describe it, the process feels similar to raising a child. First, you spend a long time teaching it words, and then one day it starts pointing at objects and naming them on its own.
On rare slow days, our annotators call several times a day asking us to speed up the delivery of new material. A remarkable program with remarkable effects.
By the way, not everyone succeeds at this work. Even during the selection stage, it becomes obvious who is patient and who is not, who gradually falls under the spell of the process, and who simply wants to make money without admitting their own annotation mistakes.
Perhaps this is already too much for a first article. Maybe there will be a continuation. Or maybe not — if this gets too much hate.
In any case, the neural network annotation software can be downloaded for free from the SpesLab website. No advertising except the manufacturer’s logo.
Olesya Grishanina
=SpesLab=
