#computer vision
Computer Vision: 34 AI articles covering computer vision news, analysis, and research
Articles
EgoHand: Monocular 3D Hand Reconstruction from Egocentric VideoNEWβ8
EgoHand reconstructs 3D hands from egocentric video in real time, trained on 14M images and achieving state-of-the-art results under the OpenMDW license.
EGGHead: Efficient Generalizable Gaussian Heads from a Single Imageβ9
EGGHead generates photorealistic, generalizable 3D Gaussian heads from a single image in one forward pass, using semantic-aware Gaussian allocation for fast, ef...
Orbia: A New Benchmark for Evaluating 3D Consistency in AIβ9
Orbia introduces a benchmark for systematically evaluating 3D consistency in AI-generated interactive video worlds, testing input preservation, content persiste...
Towards Automatic Digital Garment Assembly: A Learning-Basedβ8
Learn-based methods automate digital garment assembly from 2D patterns, predicting sewing operations and panel placement for 3D drape simulation, integrated int...
Mosqitter Is Turning Mosquito Traps Into an AI-Powered Diseaseβ9
Mosqitter's AI-powered mosquito traps create a global disease surveillance network, tracking species and population changes to warn of emerging health threats.
How Vision Transformers Represent Position: A Mechanisticβ9
A mechanistic study of Vision Transformers reveals how absolute and rotary positional encodings represent spatial position differentlyβexplicit features versus ...
A Camera-Native Stereo VR180 Dataset for Immersive Video Researchβ9
A firsthand stereo VR180 dataset with 1,211 camera-native RAW and fisheye samples, plus tools and annotations, benchmarking fisheye vs. half-equirectangular cod...
Fractal Cross Product: Theory, Differentiable Implementation andβ10
The fractal cross product generalizes the Euclidean cross product to non-integer degrees, with differentiable implementation and improved AUROC in medical image...
Spatial Lifting for Dense Prediction: A Parameter-Efficient Highβ9
Spatial Lifting lifts 2D images into higher-dimensional space for dense prediction, cutting parameters and inference costs while enabling uncertainty estimation...
STATERA: Hidden Mass Estimation via Zero-Shot Sim-to-Realβ9
STATERA adapts frozen V-JEPA video features with a temporal tubelet mixer to estimate hidden center-of-mass from monocular video, improving zero-shot sim-to-rea...