Building a Custom Image Upscaling Model
Developers are building a custom upscaling model to restore old photos locally, avoiding the limitations of commercial SaaS platforms.

Developers are building a custom upscaling model to restore old photos locally, avoiding the limitations of commercial SaaS platforms.

VisionLabs has implemented its Luna Cars computer vision system at a Moscow office of MTS to automate parking access and vehicle identification.
Building a dense point cloud using a stereo camera in ROS 2 involves frame rectification, coordinate system alignment, and fine-tuning disparity algorithms for real-time performance.

A new benchmark, Humanity's Sixth Sense, reveals that top AI models struggle to interpret visual context, achieving significantly lower accuracy than humans in intuitive reasoning tasks.

A study from IIIT Hyderabad reveals that medical AI models often rely on statistical probability rather than actual pathology when analyzing chest X-rays.

A new open-source offline tool, YOLO Annotator, offers local labeling for YOLO, Detectron2, and Mask2Former, supporting complex segmentation tasks without server infrastructure.

Sibur's engineering team upgraded a video analytics system for hundreds of cameras at a petrochemical plant without interrupting continuous production or using GPUs.

Construction companies are deploying computer vision systems based on YOLO and DETR models to automate safety monitoring and PPE detection on building sites.
The GdeKotBot project in Kazakhstan is using AI-powered cameras and GPS trackers mounted on free-roaming cats to help locate missing pets.

TwelveLabs has launched Pegasus 1.6, a video intelligence model designed to automate the labeling of egocentric footage for robotics and physical AI training.

Severstal has launched its proprietary AI-based edge inspection system, ELIS, at the Cherepovets Steel Mill, replacing foreign technology as part of a 2 ₽ billion digital initiative.
A new proof-of-concept project explores developing an AI-powered smart sight to help infantry detect and engage fast-moving drones at short range.

E-commerce search is evolving from simple image matching to multimodal models that aggregate product photos, text descriptions, and user behavior for higher accuracy.

Kordiant holding has deployed an AI-powered computer vision system at its Kaluga plant to automate quality control and prevent downtime on the rubber mixing line.
A developer built a computer vision tool to classify buckwheat quality, deploying the model as a standalone HTML file for offline use on production lines.

The new TAPe+ML v3 architecture achieves state-of-the-art performance in computer vision tasks with under 100,000 parameters, enabling efficient edge deployment and rapid video indexing.

Severstal has reported a 25% increase in the detection of hazardous workplace activities following the deployment of its Sova computer vision platform across multiple production sites.

Tevian and RSS are integrating computer vision algorithms into the Ommatid platform to enhance video analytics capabilities for large-scale infrastructure projects.

ITMO University students have created an AI prototype that automates video labeling for robot training, potentially reducing manual workload by 80 percent.

1C-Rarus has released an updated version of its AI-powered self-service terminal, cutting the hardware cost by more than half.
