3DV 2024

147 papers
The 2024 edition of the 3DV conference featured 165 papers focusing on advancements in 3D computer vision and geometric deep learning. Key research themes included 6D pose estimation, differentiable rendering for occupancy prediction, and the integration of 2D foundation models for open-vocabulary 3D perception. Notable contributions addressed the accuracy-latency trade-off in spatial refinement and explored semantic arithmetic within 3D latent spaces. Research also emphasized the decoupled optimization of geometry and appearance, alongside methods for dynamic scene reconstruction involving topological transitions. Prolific contributors included Marc Pollefeys, Michael J. Black, and Federico Tombari. Technical trends highlighted neural implicit scene encoding, as demonstrated in NICER-SLAM and RaNeuS, and self-supervision for automotive Lidar point clouds via BEVContrast. Other significant work focused on text-to-3D generation through TextMesh, as well as recovering shape and reflectance under unknown lighting. The proceedings also featured developments in model-free category-level pose estimation and experience replay for continual learning, reflecting a broad interest in both efficiency and generalizability in 3D environments.

Top Research Directions

Research directions most represented by 3DV 2024 papers.

Overcoming the Static Scene Assumption 5
Granular Structural Conditioning for 3D Generation 2
Generative Simulation for Scaling Embodied AI 2
Physics Constrained Generative Human Motion Synthesis 2
3D Reconstruction of Textureless Surfaces 2
Resolving Geometric Symmetry Ambiguity in 3D Pose Estimation 2
Accelerating Neural Scene Representations for Real Time Rendering 2
Generalizing from Synthetic Data to the Real World 2
Patch Based Spatial Context for Neural Rendering 1
Active Texture Augmentation for Stereo Correspondence 1
Continuous Level of Detail Blending in 3D Rendering 1
Predictive Spatial Hallucination for Autonomous Navigation 1
High Frequency Geometric Complexity in 3D Hair Modeling 1
Breaking the Temporal Continuity Constraint in Visual Learning 1
Mitigating Codebook Collapse in Vector Quantization 1

Top Authors

AuthorPapers
Gordon Wetzstein5
Michael J. Black5
Vladislav Golyanik5
Christian Theobalt4
Justus Thies4
Marc Pollefeys4
Daniel Cremers3
Federico Tombari3
Otmar Hilliges3
Aljaz Bozic2
Angel X Chang2
Chen Zhao2
Christian Rupprecht2
Cristian Sminchisescu2
Daniel Barath2
Deva Ramanan2
Dimitrios Tzionas2
Fabian Manhardt2
Florian Bernard2
Francesc Moreno-Noguer2

Impactful Papers

Papers from 3DV 2024 that shaped research clusters.

Foundational

Influential

Papers

Title
Pix4Point: Image Pretrained Standard Transformers for 3D Point Cloud Understanding
POCO: 3D Pose and Shape Estimation with Confidence
PRAGO: Differentiable Multi-View Pose Optimization From Objectness Detections
Purposer: Putting Human Motion Generation in Context
PU-SDF: Arbitrary-Scale Uniformly Upsampling Point Clouds via Signed Distance Functions
Q-REG: End-to-End Trainable Point Cloud Registration with Surface Curvature
Quantum-Hybrid Stereo Matching With Nonlinear Regularization and Spatial Pyramids
RadSplat: Radiance Field-Informed Gaussian Splatting for Robust Real- Time Rendering with 900+ FPS
RaNeuS: Ray-adaptive Neural Surface Reconstruction
Range-Agnostic Multi-View Depth Estimation with Keyframe Selection
Ray-Patch: An Efficient Querying for Light Field Transformers
RelPose++: Recovering 6D Poses from Sparse-view Observations
Revisiting Depth Completion from a Stereo Matching Perspective for Cross-domain Generalization
Revisiting Map Relations for Unsupervised Non-Rigid Shape Matching
RISE-SDF: A Relightable Information-Shared Signed Distance Field for Glossy Object Inverse Rendering
RIVQ-VAE: Discrete Rotation-Invariant 3D Representation Learning
ROAM: Robust and Object-Aware Motion Generation Using Neural Pose Descriptors
Robust Point Cloud Processing Through Positional Embedding
RoomDesigner: Encoding Anchor-latents for Style-consistent and Shape-compatible Indoor Scene Generation
S4C: Self-Supervised Semantic Scene Completion With Neural Fields
SALUDA: Surface-based Automotive Lidar Unsupervised Domain Adaptation
SceNeRFlow: Time-Consistent Reconstruction of General Dynamic Scenes
SCENES: Subpixel Correspondence EstimationWith Epipolar Supervision
SCENIC: Scene-aware Semantic Navigation with Instruction-guided Control
Self-Supervised Learning of Neural Implicit Feature Fields for Camera Pose Refinement
Self-Supervised Learning of Skeleton-Aware Morphological Representation for 3D Neuron Segments
Single-view 3D Scene Reconstruction with High-fidelity Shape and Texture
SlimmeRF: Slimmable Radiance Fields
Sparse 3D Reconstruction via Object-Centric Ray Sampling
SPHEAR: Spherical Head Registration for Complete Statistical 3D Modeling
Split, Merge, and Refine: Fitting Tight Bounding Boxes via Over-Segmentation and Iterative Search
Stable Surface Regularization for Fast Few-Shot NeRF
SUCRe: Leveraging Scene Structure for Underwater Color Restoration
SuperGSeg: Open-Vocabulary 3D Segmentation with Structured Super-Gaussians
TADA! Text to Animatable Digital Avatars
TECA: Text-Guided Generation and Editing of Compositional 3D Avatars
TeCH: Text-Guided Reconstruction of Lifelike Clothed Humans
Test-Time Augmentation for 3D Point Cloud Classification and Segmentation
TextMesh: Generation of Realistic 3D Meshes From Text Prompts
TRASE: Tracking-free 4D Segmentation and Editing
UAVD4L: A Large-Scale Dataset for UAV 6-DoF Localization
UniGaussian: Driving Scene Reconstruction from Multiple Camera Models via Unified Gaussian Representations
UniMotion: Unifying 3D Human Motion Synthesis and Understanding
Unsupervised 3D Keypoint Discovery with Multi-View Geometry
Unsupervised Representation Learning for Diverse Deformable Shape Collections
Zero-BEV: Zero-shot Projection of Any First-Person Modality to BEV Maps
ZeroHSI: Zero-Shot 4D Human-Scene Interaction by Video Generation

« prev Page 3/3