Computer Vision news and analysis on Web Pulse: 3338 curated articles tracking the latest Computer Vision developments, tools, and research, updated continuously from vetted sources.
Researchers propose Uncertainty-Guided Cascade Forward Refinement (U-CFR), a new inference-time framework for interactive image segmentation that enables models to autonomously self-correct after each user interaction. U…
Researchers present ODeform, a method extending Neural Ordinary Differential Equations to model continuous 4D dynamics of deformable objects in 3D space, eliminating discrete time steps while maintaining computational ef…
Researchers from the Data Science and Analytics Group at Georgia Tech (DS@GT ARC) developed a privacy-preserving framework for synthetic lung CT slice generation for the ImageCLEFmed GANs 2026 challenge, achieving a Priv…
Researchers introduced ViSTR-Bench, a benchmark to evaluate whether multimodal large language models (MLLMs) can reason from continuous visual cues in dynamic scenes, finding that current models remain far below human pe…
Axolotl3D, a multi-modal and occlusion-aware 3D generation model introduced in arXiv:2607.20660v1, jointly conditions on images, visibility masks, camera parameters, and a partial point cloud to enable faithful 3D shape …
Researchers introduced Lumera, a benchmark and pipeline for engine-native, light-aware 3D scene parsing from a single image, built from 2,513 UE5 projects with 3.73M components and 102.6K parametric lights. Lumera-Box ac…
A new study evaluating iris recognition on VR-acquired data finds that standard ISO/IEC 29794-6 quality metrics, such as margin adequacy, fail on VR-acquired data, while multimodal fusion of iris and periocular recogniti…
Researchers have developed a zero-shot Digital Twin framework that uses a Thermodynamics-Informed Graph Neural Network to simulate physical systems on unseen geometries without retraining, operating at approximately 25 m…
Researchers have introduced RealVDeblur, an efficient generative framework for real-world video deblurring that uses a physically grounded blur synthesis pipeline from 3D Gaussian Splatting assets and high-frame-rate vid…
Researchers present 3D-GIMP (3D Gaussian Inpainting Meets Patch Matching), a hybrid paradigm for high-fidelity object removal in 3D Gaussian Splatting that performs a single generative inpainting on a key reference view …
Developer danilo-znamerovszkij has open-sourced a pipeline that converts handwriting photos into TTF fonts using npm, with a Claude Code skill wrapper that automates vision labeling and local build. The deterministic ima…
A study by researchers at the MIT Media Lab, including Wolf, E., Samaradivakara, Y., Gokhale, O., Ahmed, S., Wang, Y., Pataranutaporn, P., and Maes, P., published in the 2026 CHI Conference Extended Abstracts, found that…
A research-led guide compares top multimodal embedding models for image-text retrieval, visual documents, video, audio, and multimodal RAG, emphasizing that retrieval systems must preserve fine-grained details lost durin…
A developer built iTube, a pipeline that turns entire novels into narrated, lip-synced motion videos on a single 16GB GPU using FLUX, Wan2.2, PuLID, MuseTalk, and ComfyUI. The key insight is that audio length dictates sc…
Deloitte projects generative AI fraud losses will hit $40 billion in the United States by 2027, up from $12.3 billion in 2023, and a joint policy paper from the American Bankers Association, the Better Identity Coalition…
Researchers upgraded a deep learning model using 1960s CORONA satellite imagery to automatically detect archaeological sites in Mesopotamia, achieving over 85% Intersection over Union and 90% detection accuracy in the Ab…
Researchers have developed a Deformable Detection Transformer (DETR)-based object detection model that achieved up to 28-37% improvement over prior work on COCO-style metrics for automated cuneiform sign detection, using…
A new deep learning model called Q2000, developed by researchers using G-LiHT Lidar data, detects ancient Maya settlements across 35,584 km2 of the Yucatan Peninsula with an F1 score of 0.89, matching the accuracy of sma…
Team Hope's ResNet34 model, fine-tuned with ImageNet pretrained weights for 25 epochs using the Adam optimizer and a learning rate of 1e-4, achieved 81% accuracy on the NakbaArchiveClassifier Shared Task at Nakba-NLP 202…
Velmeni received FDA 510(k) clearance for VELMENI for DENTISTS 3D, AI-assisted software that segments anatomy and reconstructs views from cone-beam computed tomography scans. The clearance gives CEO Mini Suri's company a…