3DV 2021
110 papers
The 2021 edition of the International Conference on 3D Vision (3DV) featured 163 papers focusing on spatial intelligence and geometric modeling. Research themes centered on human-centric 3D vision, including perspective-aware pose estimation, template-free reconstruction of articulated objects, and neural occupancy representations for hands. Depth estimation was a prominent area of study, with work addressing temporal consistency in video and geometry-guided attention for self-supervised monocular depth learning.
Methodological trends included coarse-to-fine refinement strategies for dense correspondence and the development of category-agnostic priors for universal 3D understanding. Data augmentation techniques, such as Mix3D for out-of-context scene processing, and advancements in neural radiance fields (NeRF) for object category modeling were also featured. Prolific contributors included Otmar Hilliges, Marc Pollefeys, and Luc Van Gool, among others. Additional research explored 3D surface super-resolution and occlusion-aware geometric consistency, reflecting a shift toward robust, self-supervised learning frameworks in 3D environments.
Top Research Directions
Research directions most represented by 3DV 2021 papers.
Top Authors
Impactful Papers
Papers from 3DV 2021 that shaped research clusters.
Foundational
Neural Disparity Refinement for Arbitrary Resolution Stereo
foundational
Stereo Matching for Asymmetric and Unbal
TermiNeRF: Ray Termination Prediction for Efficient Neural Rendering
foundational
Surface Constrained Neural Rendering
RefRec: Pseudo-labels Refinement via Shape Reconstruction for Unsupervised 3D Domain Adaptation
foundational
Geometric Invariance for 3D Point Cloud
TermiNeRF: Ray Termination Prediction for Efficient Neural Rendering
foundational
Baking Neural Radiance Fields into Textu
Influential
RefRec: Pseudo-labels Refinement via Shape Reconstruction for Unsupervised 3D Domain Adaptation
influential
Refining Labels via Local Feature Space
Multi-Category Mesh Reconstruction From Image Collections
influential
Template Free 3D Reconstruction of Gener
PanoDepth: A Two-Stage Approach for Monocular Omnidirectional Depth Estimation
influential
Environmental Generalization in 3D Scene
TermiNeRF: Ray Termination Prediction for Efficient Neural Rendering
influential
Accelerating Neural Scene Representation
A Spatio-temporal Transformer for 3D Human Motion Prediction
influential
Spatio-Temporal Attention for Skeleton a
NeeDrop: Self-supervised Shape Representation from Sparse Point Clouds using Needle Dropping
influential
Reconstructing Complete Geometry from Pa
Joint 3D Human Shape Recovery and Pose Estimation from a Single Image with Bilayer Graph
influential
Resolving Self Occlusion in Articulated
A Spatio-temporal Transformer for 3D Human Motion Prediction
influential
Disentangling Space and Time in Motion F
Task-Generic Hierarchical Human Motion Prior using VAEs
influential
Composite Loss Functions for Perceptual
Papers
« prev Page 3/3