Ultralytics has introduced Ultralytics Agents, a visual workflow builder for computer vision. It connects YOLO models, vision-language models, conditions, datasets, and actions. Teams can automate alerts, collect uncertain detections, and trigger downstream actions. The platform also creates a feedback loop for reviewing, annotating, retraining, and redeploying models. [link]

Ambarella and Ultralytics announced a collaboration to bring Ultralytics YOLO models to Ambarella’s low-power CVflow edge AI devices. The integration targets real-time detection, segmentation, and tracking across smart cameras, robotics, automotive, and industrial applications, enabling lower latency, reduced bandwidth needs, and less reliance on cloud processing. [link]

UVeye and LaFontaine Automotive Group are creating Michigan’s first statewide AI vehicle-inspection network, installing UVeye’s drive-through systems across all 43 dealerships. The computer-vision technology scans tires, undercarriages, and exteriors for defects in seconds, with UVeye claiming 96% accuracy. Scans will be free for vehicles of any make or model. [link]

C5MI and Slamcore announced a strategic partnership to bring infrastructure-free RTLS to material-handling fleets across North America. C5MI will resell Slamcore Aware and Hub, combining vision-based localization with SAP EWM, KINEXON OS and Omnipoint. The solution provides real-time MHE tracking without fixed infrastructure, targeting warehouses, factories, defense logistics, safety and productivity. [link].

Buildots raised $130 million, led by O.G. Venture Partners, bringing total funding to $297 million. The construction AI company will expand enterprise deployments, cover more lifecycle stages, and turn portfolio data into insights using computer vision and digital twins built from real construction sites. [link]

United Airlines Ventures invested in Assaia to support StandManager, an AI-powered stand and gate management solution. Part of Assaia’s ResourceManager platform, it combines operational data, business rules, computer vision and real-time conditions to improve airport gate allocation and operational decision-making. [link]

Aerospace visual inspection faces limits: it can slow production, miss subsurface defects, and vary with defect type, severity, viewing angle, lighting, glare and access. NASA, FAA and research highlight matching inspection methods to specific defects and conditions.

Yes! Apples has joined Clairfresh to centralize quality management across operations. The partnership aims to improve consistency, visibility and standardization while creating a scalable quality foundation that aligns teams, streamlines processes and supports the company’s continued growth. [link]

Researchers introduced MambaMPD, a Mamba-based segmentation framework for detecting marine pollution from satellite imagery. Designed for challenging conditions such as small pollution patches, low contrast, fragmented patterns, and unclear boundaries, it combines Frequency-Aware Augmentation (FAA) using wavelet transforms with Edge-Guided Attention (EGA) for sharper boundary detection. [link]

ModaLens introduces a paired image-swap audit to test whether medical vision-language models actually respond to radiology images when reports are available. Evaluating MedGemma-27B on 44,786 trials from 3,199 MIMIC-CXR cases, researchers found that generated answers changed after image substitution in 20.94% of trials without the report, compared with just 4.26% with it—a 16.7-point difference. [link]

In a recent article in CTech by Calcalist, Amir Hever, CEO of UVeye discusses the overlooked role of vehicle condition, inspection, and real-time physical data in making autonomous driving safer and reliable

In the recent Retail Tech podcast, Lumana’s Orit Dolev explains how AI-powered video intelligence transforms existing retail cameras into real-time systems, detecting shrink, monitoring shelves, managing queues, optimizing workforce allocation, and improving store operations.

A latest Reddit post discusses about the latest model PP-OCRv6 that can be used for OCR tasks across industries.

Another reddit post discusses on training with unbalanced classes and data scarcity.