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.
12 day(s) with sentiment data
-
New Local-Frame Rotary Attention method for antibody epitope prediction
Researchers have introduced Local-Frame Rotary Attention (LFRA), a novel mechanism for predicting antibody-specific epitopes. This method directly incorporates three-dimensional residue geometry into the attention mecha…
-
New SO(3)-equivariant GNN processes CAD boundary representations
Researchers have developed SO3brep, a novel SO(3)-equivariant graph neural network designed to process CAD boundary representations (B-reps). This new model addresses the challenge of learning from CAD models that are n…
-
CalibSwitch framework enhances camera-LiDAR sensor calibration for multimodal perception
Researchers have developed CalibSwitch, a novel framework designed to improve the accuracy and reliability of camera-LiDAR sensor calibration for multimodal perception systems. The system separates the process into abso…
-
Rotation-equivariant deep learning improves brain age prediction
Researchers have developed a novel framework for predicting brain age using diffusion MRI data, which captures white matter fiber orientation distributions. This approach utilizes a rotation-equivariant Convolutional Ne…
-
RADIANT architecture offers faster, object-centric global illumination rendering
A new object-centric architecture called RADIANT has been developed for feed-forward global illumination, addressing the quadratic scaling issues of previous transformer-based renderers. RADIANT represents each object w…
-
New IsoAct Framework Enhances Geometric Debiasing for AI Representations
Researchers have introduced IsoAct, a novel post-hoc framework designed to debias machine learning representations, particularly those residing on non-Euclidean manifolds. Unlike previous methods that assume Euclidean s…
-
HiP3R method enhances 3D reconstruction from long videos
Researchers have developed HiP3R, a novel hierarchical pose graph method designed to improve the accuracy and consistency of 3D reconstructions from long video sequences. Traditional methods struggle with the computatio…
-
New REFLEX method generates diverse RNA ensembles with high fidelity
Researchers have developed REFLEX, a new method for generating RNA ensembles that accurately captures diverse conformational states without collapsing to averaged structures or drifting into unsupported regions. This se…
-
New pipeline generates thousands of robot demonstrations from single human video
Researchers have developed Real2Sem2Real (R2Sem2R), a novel pipeline that transforms a single human RGB video demonstration into thousands of robot demonstrations. This method utilizes DINOv3-distilled Gaussian splats t…
-
VoxProg introduces voxel programming for 3D spatial reasoning
Researchers have introduced VoxProg, a novel approach to 3D spatial reasoning that utilizes voxel programming. This method enables language models to translate task interpretations into executable programs that operate …
-
HiP3R method improves 3D reconstruction with hierarchical pose graphs
Researchers have developed HiP3R, a novel method for long-horizon 3D reconstruction that addresses the challenges of merging local video reconstructions into globally consistent camera trajectories and 3D geometry. Unli…
-
EMERGE: New SE(3)-equivariant diffusion model for point cloud generation
Researchers have developed EMERGE, a novel graph-based diffusion model for generating point clouds. This model is the first to be fully SE(3)-equivariant, meaning it preserves continuous spatial symmetries and avoids ri…
-
AI Scaling Laws, Scalable Systems, and AI-Generated Film Production Explored
A technical note published on arXiv delves into the foundational principles of scaling laws in AI, explaining the relationships between model size, training data, and compute resources through geometric and spectral ana…
-
New HydroShear simulator boosts tactile sim-to-real transfer for robotics
Researchers have developed HydroShear, a novel hydroelastic simulator designed to improve the accuracy of tactile feedback in reinforcement learning for robotics. This simulator models complex phenomena like stick-slip …
-
EMBER framework predicts TCR-pMHC binding without complex docking
Researchers have developed EMBER, a novel framework for predicting T-cell receptor (TCR) and peptide-MHC (pMHC) binding. Unlike traditional methods that require a docked complex, EMBER analyzes independently folded TCR …
-
New cuVSLAM system enhances robot navigation with line features
Researchers have developed a new visual odometry system called Line-Augmented cuVSLAM, designed to improve robot navigation in environments with low texture. This system extends the existing cuVSLAM by incorporating lin…
-
New method optimizes SE(3)-equivariant structure prediction with faster kNN graphs
Researchers have developed a novel approach to optimize topology-pass scheduling in SE(3)-equivariant structure prediction. This method addresses the computational expense of rebuilding k-nearest neighbour (kNN) graphs …
-
NeuSE system enhances object-based SLAM with SE(3)-equivariant embeddings
Researchers have developed NeuSE, a novel neural embedding system designed for object-based Simultaneous Localization and Mapping (SLAM). This system creates latent object embeddings that capture an object's shape, scal…
-
New Phase-Conditioned Affordance Field Boosts Robot Visuomotor Policy Performance
Researchers have developed a new method called Phase-Conditioned Affordance Field (PAF) to improve the performance of visuomotor policies in robotics. This approach introduces a local geometric inductive bias by learnin…
-
New SIEVE framework tackles topological obstruction in 3D point cloud learning
Researchers have developed SIEVE, a new SE(3)-equivariant self-supervised learning framework designed to overcome a theoretical incompatibility between 3D point cloud masked autoencoding and SE(3)-equivariant networks. …