Computer vision is one of our deepest areas, covering how machines learn to see, segment, and reason about images and video. Articles here range from convolutional and transformer based architectures to dense prediction tasks like detection and segmentation, with regular coverage of medical imaging where reliable vision models carry real clinical weight. The emphasis stays on what makes a method work and where it breaks, backed by the original research.
AngioGraphCAD models coronary arteries as graphs, letting a graph neural network assess lesion-level risk the way cardiologists reason — one lesion at a time.
A team from Balgrist University Hospital and ETH Zurich built an instance-specific self-supervised framework that registers bone surfaces across imaging modalities by learning a continuous neural distance field from the preoperative scan and using thousands of neural hypothesis generation heads…
A team from ShanghaiTech University, Henan Provincial People’s Hospital, and Shanghai United Imaging Intelligence built a hierarchical framework that classifies 23 brain disorders across three diagnostic levels, cuts the performance gap between coarse and fine-grained predictions to just 2%, and…
A team from Wuhan University of Science and Technology, Xiamen University, and the University of New South Wales built a multiple-magnification MIL framework that uses optimal transport to match tissue features across magnification levels, outperforming every tested baseline across four…
Strip away the buzzwords and a computer vision pipeline is a chain of functions: raw image in, structured data out. Each function transforms the image or extracts information from it. The hard part isn’t any single step — it’s connecting…
Building Computer Vision Pipelines with Claude Code (2026 Guide) Read More »
YOLO (You Only Look Once) has been the practical standard for real-time object detection since 2015. What started as a single-pass detector that traded some accuracy for dramatic speed has become a mature family of models, each generation narrowing the…
YOLOv11 Object Detection: From Zero to Deployment (2026 Guide) Read More »
Tianyi Lin, Chi Jin, and Michael I. Jordan from Columbia, Princeton, and UC Berkeley deliver the first rigorous nonasymptotic analysis of two-timescale gradient descent ascent — the algorithm that has powered GAN training in practice for years — establishing tight…
TTGDA: Two-Timescale Gradient Descent Ascent for Nonconvex Minimax Optimization Read More »
A multi-institution team from UC Berkeley, Stanford, Meta AI, and UC San Diego shows that training a vision model on the last 1.6 seconds of video — and nothing else — consistently beats both the fixed-model baseline and an offline…
Test-Time Training on Video Streams: Why Forgetting Is Actually a Feature Read More »
A multi-institution team spanning Peking University, UC San Diego, Stony Brook University, and Weill Cornell Medicine built PEGN — a framework that injects localized topological features from extended persistent homology into graph neural networks, pushing expressiveness beyond the 3-WL barrier…
PEGN: How Persistent Homology Breaks the WL Barrier in Graph Neural Networks Read More »
When you only have a handful of real sinkhole examples, you either accept poor detection or get creative about generating training data. Researchers at Arts et Métiers ParisTech and SNCF Réseau chose the latter — building physics-informed synthetic sinkholes, embedding…