3D Vision

LoCo: Memory-efficient location-consistent image features learned via a novel ranking loss, enabling three orders of magnitude memory improvement and outperforming state-of-the-art.

LinNet: A linear-time point cloud network achieving 10x speedup over PointNeXt, with state-of-the-art accuracy on various benchmarks.

LE3D: Real-time HDR view synthesis from noisy RAW images is achieved using 3DGS, significantly reducing training time and improving rendering speed.

LightGaussian achieves 15x compression of 3D Gaussian scene representations, boosting rendering speed to 200+ FPS while maintaining visual quality, solving storage and efficiency issues in real-time n…

Learned Mixture Density Particle Smoother (MDPS) surpasses state-of-the-art for accurate, differentiable city-scale vehicle localization.

Create animatable interacting hand avatars from a single image using a novel two-stage interaction-aware 3D Gaussian splatting framework!

DisPA: a novel disentangled representation learning framework for perceptual point cloud quality assessment achieves superior performance by minimizing mutual information between content and distortio…

Animates diverse garments realistically from a single cloth’s trajectory using a disentangled learning approach and Energy Unit Network (EUNet).

3D surface reconstruction revolutionized: PEIF leverages patch-level pose-invariant representations and 3D patch-level equivariance for state-of-the-art accuracy, even with varied poses and datasets!

LCM: a novel, locally constrained, compact point cloud model surpasses Transformer-based methods by significantly improving performance and efficiency in various downstream tasks.

Large Spatial Model (LSM) achieves real-time semantic 3D reconstruction from just two unposed images, unifying multiple 3D vision tasks in a single feed-forward pass.

LAM3D: A novel framework uses point cloud data to boost single-image 3D mesh reconstruction accuracy, achieving state-of-the-art results in just 6 seconds.

L4GM: The first 4D model generating high-quality animated 3D objects from single-view videos in a single feed-forward pass.

Key-Grid: An unsupervised 3D keypoint detector achieving state-of-the-art semantic consistency and accuracy for both rigid and deformable objects using novel grid heatmap features.

This research presents LocalN2NM, a novel method for inferring neural signed distance functions (SDF) from single, noisy point clouds by finetuning data-driven priors, achieving faster inference and b…

Meta-learning enhances human mesh recovery by unifying training and test-time objectives, significantly improving accuracy and generalization.

In-N-Out: Lifting 2D Diffusion Priors for 3D Object Removal via Tuning-Free Latents Alignment enhances 3D scene reconstruction by aligning 2D diffusion model latents for consistent multi-view inpainti…

Boosting 3D point cloud recognition robustness, Frequency Adversarial Training (FAT) leverages frequency-domain adversarial examples to improve model resilience against corruptions, achieving state-of…

ImOV3D: Revolutionizing open-vocabulary 3D object detection by learning from 2D images alone!

IllumiNeRF: Relightable 3D reconstruction without inverse rendering using image diffusion and NeRF.