#computer vision
Computer Vision: 34 AI articles covering computer vision news, analysis, and research
Articles
Multi-Task Morphological Concept Learning for Pulmonary Nodule⭐8
A study evaluates multi-task 3D CNN learning of radiologist-annotated morphological features for pulmonary nodule malignancy assessment on LIDC-IDRI CT scans.
AlphaEarth Distinguishes Cities but Compresses Urban Variation⭐9
AlphaEarth embeddings distinguish cities globally but compress urban variation, with dispersion linked to national development and vegetation contrast.
UltraBench 2: A New Benchmark for Evaluating Vision Foundation⭐8
UltraBench 2 benchmarks vision foundation models on ultrasound imaging, showing domain-specific pretraining leads in classification while general models match s...
Token Clustering and Semantic Sequence Mamba for Hyperspectral⭐9
STMamba uses token clustering and semantic sequence Mamba to organize sparse tokens into coherent sequences, improving hyperspectral image classification accura...
Physiology-Aware Gaussian Head Avatar Embeds Recoverable Cardiac⭐8
Heartian embeds cardiac signals into Gaussian head avatars for realistic rPPG synthesis, achieving 0.29 bpm heart-rate MAE while preserving reconstruction quali...
Did You Steal My Shot? Pioneering Camera Motion Plagiarism⭐9
Researchers propose the first camera motion plagiarism detection benchmark for generative videos, using optical flow and vorticity cues to spot stolen cinematog...
RAUL: Reference-Assisted Ureteroscopy Localization for Skill⭐9
RAUL recovers ureteroscope trajectories from video alone—no tracking hardware—achieving 0.5 mm accuracy and 86% coverage for objective surgical skill assessment...
MechReason: A New Benchmark for Multi-Image, Multi-Hop Reasoning⭐8
MechReason, a 12k-question benchmark from real mechanical engineering papers, tests multi-image, multi-hop reasoning in MLLMs across explanation, prediction, de...
Feature Recovery for Object Understanding After Irreversible⭐7
TRACE benchmark and Feature Recovery Module improve post-fire object detection and identification when irreversible damage degrades vision model performance.
MHE-Former: Multi-Hypothesis Transformers via Entropy⭐8
MHE-Former uses entropy-maximized multi-hypothesis Transformers and VLM-based selection to recover 3D hand and body meshes from monocular images with state-of-t...