SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks
PulseAugur coverage of SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks — every cluster mentioning SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks across labs, papers, and developer communities, ranked by signal.
9 day(s) with sentiment data
-
New differential pose estimation method enhances robotic motion accuracy
Researchers have developed a novel differential pose estimation method designed to improve accuracy and robustness in 6-DOF motion estimation for robotics and autonomous systems. This new approach directly calculates pl…
-
New Polyatomic Complexes representation for atomistic systems unveiled
Researchers have developed a new learning representation for atomistic systems called Polyatomic Complexes, designed to be invariant to physical symmetries and unique. This representation, detailed in a paper submitted …
-
New neural network framework learns complex Lie-Poisson system dynamics
Researchers have developed Latent Lie-Poisson Neural Networks (LLPNNs), a novel framework designed to learn and predict the dynamics of Lie-Poisson systems directly from observable data. These systems are crucial for mo…
-
New pipeline generates infinite 3D worlds from single panoramas
Researchers have developed Genie Sim PanoWorld, a novel pipeline for generating infinite indoor 3D worlds from a single panoramic image. This system bypasses the need for per-scene optimization or multi-view capture, ad…
-
SM4RT Transformer learns structured motion geometry for 4D reconstruction
Researchers have developed SM4RT, a novel Transformer-based model designed for 4D reconstruction and structured motion perception from monocular RGB video. Unlike previous methods that treat motion as independent point-…
-
TriGlue: Biology-Inspired Generative Model for Molecular Glue Design
Researchers have introduced TriGlue, a novel generative model inspired by biological mechanisms for designing molecular glues. This model addresses the complex challenge of designing molecular glues, which requires simu…
-
New deep learning framework OrbitAll unifies molecular system representation
Researchers have developed OrbitAll, a novel deep learning framework designed to represent all molecular systems using quantum mechanical principles. This framework integrates spin-polarized orbital features with SE(3)-…
-
New InertialAR model advances 3D molecule generation with Transformer++
Researchers have developed InertialAR, a novel autoregressive model for generating 3D molecules. This model addresses key challenges in molecule tokenization by creating an SE(3)- and permutation-invariant sequence of t…
-
DyneTrion emulator models protein dynamics across timescales
Researchers have developed DyneTrion, a novel generative emulator designed to model protein dynamics across various timescales. This system utilizes a tri-attention architecture that combines SE(3)-robust geometric upda…
-
New SeeSE3 method reveals 3D space understanding in vision models
Researchers have developed a new method called SeeSE3 to investigate whether vision foundation models inherently understand 3D Euclidean space. Unlike prior approaches that focus on predicting 3D properties like depth, …
-
GenVid2Robot framework translates generated video motion into executable robot trajectories
Researchers have developed GenVid2Robot, a framework that translates generated video motion into executable robot manipulation trajectories. This system addresses the limitations of using generated videos directly for r…
-
RayRoPE introduces projective positional encoding for multi-view 3D attention
Researchers have developed RayRoPE, a novel positional encoding method designed for multi-view transformers in 3D computer graphics. This new approach uniquely encodes image patches, enables SE(3)-invariant attention, a…
-
New Geometric Observability Index Enhances SE(3) Pose Estimation
Researchers have introduced the Geometric Observability Index (GOI), a novel metric for assessing the sensitivity of pose estimation in SE(3) environments. This index quantifies the influence of individual measurements …
-
VulcanVoxel learns 3D affordances for robotic blade insertion
Researchers have developed VulcanVoxel, a novel approach for robots to learn 3D affordances, specifically for blade insertion tasks in cluttered environments. Unlike traditional methods that infer SE(3) pose distributio…
-
LIME system learns intent-aware camera motion from egocentric video
Researchers have developed LIME, a novel system designed to generate intent-aware camera motion from egocentric video. LIME addresses the challenge of predicting optimal camera poses based on natural language intents, a…
-
New framework unifies geometry-preserving neural architectures on manifolds
Researchers have developed a unified framework for geometry-preserving neural architectures, organizing them based on where and how geometric constraints are enforced. This work addresses theoretical gaps by proving app…
-
Diffeomorphic optimization method enhances protein design and data manifold learning
Researchers have introduced a novel method called diffeomorphic optimization, designed to improve the process of optimizing objectives on complex data manifolds. This technique leverages diffusion and flow models to map…
-
New AI method predicts molecular Hessians 100x faster
Researchers have developed a novel method called Hessian Interatomic Potentials (HIP) that can predict molecular Hessians, crucial for computational chemistry tasks, without requiring derivative calculations. This deep …
-
CoStream framework composes simple behaviors for complex robotic manipulation
Researchers have introduced CoStream, a novel framework designed to enable complex manipulation tasks in robotics by composing simpler, independent behaviors. This approach leverages foundation models and diverse sensin…
-
Calousel method calibrates non-overlapping multi-camera systems using rotation
Researchers have developed a new method called Calousel for calibrating multi-camera systems that do not have overlapping fields of view. This approach utilizes pure rotational motion and a single static calibration boa…