Embedded Files
GAMBIT: Software for Neural Network Markup
This New Year, from 2024 to 2025, SpecLab is gifting the world a new, incredibly useful program that allows individuals to create neural networks for themselves and others. The entire process is designed for beginners, making it possible for even homemakers to create artificial intelligence—just as the great Lenin once envisioned (well, he actually envisioned something else, but it applies here too).
Yes, there are already many large companies that sell neural networks capable of recognizing any object. However, neither Google, Samsung, nor Microsoft will specially train a neural network just for you—one that lets your specific dog into your house, records the readings of your exact meter, monitors the needle on your specific instrument, checks for blockages in the storm drains around your house (or your basement will flood), tracks the quality of products in your particular factory, identifies defects on your specific machine, or ensures your employees follow the rules of your specific workplace... This list could easily be extended to a million examples—there are that many personal tasks people have in the field of AI.
But not everyone has the budget for top-tier programmers. Not everyone even understands how it works.
SpecLab provides a simple, user-friendly tool that is accessible to anyone. And you can try it right now.
To do so, simply record a video of the object or process you wish to monitor using a regular video camera or mobile phone. Then, upload the video into the “Gambit” program and highlight the objects you need. After processing a certain number of frames, you’ll obtain what’s called a DATASET, which forms the basis for training a neural network.
Upload it to our AI Cloud, where it transforms… where it transforms… into a fully functional neural network, ready for use, along with service functions. For example, the system can send you an email if your favorite plant starts wilting, or trigger a relay in your Smart Home system if the camera detects a puddle of water on the floor. Or… a million other variations—this is a real factory for neural networks.
However, this program was not originally designed for such basic applications. With it, you can create datasets for detecting serious processes and a wide variety of objects. For example, abandoned items can have hundreds of thousands of different types, from bags and packages to boxes and suitcases. SpecLab’s software allows you to offload work for others, including for SpecLab’s own needs. Interested parties—Welcome!
In this case, of course, it's not as simple as it sounds. Attention to detail and patience are required to accurately select the objects in large quantities. These qualities are not related to formal education or academic knowledge—this is more suited for people who can handle repetitive tasks, like knitting, or other tasks requiring consistent, repetitive work. Therefore, we don't accept everyone—only those who make no mistakes. But you can try; we give you three attempts. Due to the liberalization of laws, we also accept candidates from 14 years old. However, in all cases, candidates should have basic knowledge, like how to send an email, download a file, or upload it to the cloud. We will not teach you how to tie shoelaces.
Gambit is not so much a commercial product as a companion tool. We ourselves are constantly training neural networks and haven’t found anything convenient or accessible online. You can compare it with tools like AI-Markup, VisionAnnotator, ObjectMapper, SegmentPro, and SegmentMaster—all of which are not only inconvenient but also hopelessly outdated. They lack feedback, self-learning, and other useful features. SpecLab created this program primarily for internal use and applies it to many of our products. This annotation system is continuously tested by our staff, with a new version released nearly every week—smarter and more convenient each time.
We’d love to hear feedback and suggestions for new ideas in this field, because we need this too. The core functionality is free—feel free to use it. Only high-level features, which we’ve already started to integrate for neural network training, will be paid. One of them is already available, and it eliminates false positives. We haven’t seen anything like this in Google software or other AI giants. If their neural network messes up, there's nothing they can do about it. But it’s the false positives that ruin any good idea.
The quality of the dataset is what determines how effectively the neural network will perform. SpecLab has invested a great deal of experience into the methodology for creating metadata. This is explained in the help section and training videos. We also hold training seminars for anyone interested. It’s all simple and accessible—no more vague complexities in AI!
Page updated
Google Sites
Report abuse