optical flow
PulseAugur coverage of optical flow — every cluster mentioning optical flow across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
New algorithm MoViD uses photonic principles for real-time video motion detection
A new algorithm called MoViD (Motion estimation via Virtual dispersion and phase Detection) has been developed, inspired by photonic time stretch technology. This physics-based approach treats video pixels as optical ti…
-
New benchmark released for evaluating deformable 3D reconstruction
Researchers have introduced the Synthetic Deformable Tubular Benchmark, a new dataset designed to evaluate deformable 3D reconstruction. Generated using the LumenMorph toolkit, this benchmark includes 5,580 RGB images a…
-
Hybrid framework refines structure-from-motion with pose graph optimization
Researchers have developed a hybrid framework that enhances the ACE-Zero learning-based method for structure-from-motion tasks. This new approach integrates pose graph optimization (PGO) to refine camera poses and scene…
-
WorldForge V2 enhances camera control in video diffusion models
Researchers have developed WorldForge V2, a novel framework designed to enhance camera trajectory control in video diffusion models. This method employs a training-free guidance approach that refines generations along a…
-
Spiking Neural Network Framework Achieves State-of-the-Art Optical Flow
Researchers have developed SICAF, a novel framework for self-supervised event-based optical flow estimation using spiking neural networks. This approach addresses challenges in extracting spatiotemporal features from ir…
-
New system uses codec motion vectors for faster online action detection
Researchers have developed CodecMotion-OAD, a novel system for online action detection in videos that utilizes codec motion vectors (MVs) instead of traditional optical flow. This approach significantly reduces computat…
-
New CoDy framework uses video codec dynamics for unsupervised action localization
Researchers have introduced CoDy, a novel framework for unsupervised temporal action localization in videos. CoDy leverages the inherent dynamics within video codec quantization patterns, which are shown to preserve cat…
-
AI models compared for wingbeat counting in flapping-wing vehicles
Researchers have evaluated three types of temporal models—convolutional, spiking, and attention-based—for counting wingbeats in flapping-wing vehicles using optical flow data. The study, conducted in the MuJoCo simulati…
-
Review paper categorizes HDR imaging methods
This review paper categorizes and analyzes existing literature on multi-exposure High Dynamic Range (HDR) imaging. It focuses on two key areas: multi-exposure fusion (MEF) and ghost removal techniques. The paper examine…
-
Vision-based tactile sensors questioned for embodied AI due to limitations
A recent analysis questions the viability of vision-based tactile sensors (VBTS) as a foundational technology for embodied AI. The article argues that while popular, VBTS relies heavily on borrowed visual algorithms and…
-
AI advances tackle video generation consistency and speed
Recent research in AI-driven video generation focuses on improving consistency and efficiency. ContextAnyone addresses character consistency by treating reference images as explicit anchors within a shared diffusion tra…
-
PoseOFF improves human action anticipation for robots · 2 sources tracked
Researchers have developed PoseOFF, a novel pose-anchored optical flow representation designed to improve human action anticipation in human-robot interaction. This method captures localized motion around human joints, …
-
New method integrates aleatoric and epistemic uncertainty in deep learning
Researchers have developed a new method for uncertainty quantification (UQ) in deep learning models, particularly for high-dimensional output spaces. This approach jointly models aleatoric uncertainty, which accounts fo…
-
Gradient energy tensor offers faster, real-time image enhancement
A new research paper proposes the gradient energy tensor as a viable alternative to the structure tensor for various image processing tasks. The study demonstrates its application in corner detection, optical flow estim…
-
New method improves sign language recognition using temporal alignment
Researchers have developed a novel approach to sign language recognition by employing transfer learning and a domain adaptation method called TA3N. This method utilizes a Temporal Relational Network (TRN) to align multi…
-
3D Fire Reconstruction Achieved with Gaussian Splatting and Optical Flow
Researchers have developed a novel method for reconstructing dynamic fire in 3D using a spatiotemporal representation based on Gaussian functions. This technique aims to capture the complex, high-frequency features of f…
-
Video-based system counts Parkinson's steps without wearables · 2 sources tracked
Researchers have developed a novel video-based framework to passively count steps for individuals with Parkinson's disease, addressing limitations of current wearable-based methods. The system utilizes 3D human mesh rec…
-
New SAC^2-Net advances micro-expression recognition with semantic anchoring
Researchers have developed SAC^2-Net, a novel network designed to improve micro-expression recognition (MER) by addressing challenges like subtle facial movements and limited data. The system leverages the complementary…
-
New Triangular Consistency Method Enhances Optical Flow Learning
Researchers have introduced a new method called triangular consistency for learning optical flow. This technique acts as a universal constraint, applicable across various network architectures, supervision types, and da…
-
New framework enhances infrared small target detection in dynamic scenes
Researchers have developed a new framework for detecting small infrared targets in dynamic scenes, addressing the challenge of coupled motions between targets, imaging platforms, and backgrounds. The proposed method int…