liruihui / awsome_deep_geometry_learning

A list of resources about deep learning solutions on 3D shape processing

A curated list of resources for Deep Geometry Learning

Image-based methods

• 2012-NIPS - Convolutional-recursive deep learning for 3d object classification. [Paper]

• 2014-NIPS - Depth map prediction from a single image using a multi-scale deep network. [Paper]

• 2014-ECCV - Learning Rich Features from RGB-D Images for Object Detection and Segmentation. [Paper]

• 2015-CVPR - Aligning 3D models to RGB-D images of cluttered scenes. [Paper]

• 2015-ICCV - Multi-view convolutional neural networks for 3d shape recognition. [Paper]

• 2016-CVPR - Deep Sliding Shapes for Amodal 3D Object Detection in RGB-D Images. [Paper]

• 2016-CVPR - Volumetric and multi-view cnns for object classification on 3d data. [Paper]

• 2020-CVPR - Autolabeling 3D Objects with Differentiable Rendering of SDF Shape Priors. [Paper][Code]

Voxel-based methods

Dense Voxels

• 2015-CVPR - 3d shapenets: A deep representation for volumetric shapes. [Paper]

• 2015-ICCV - Multi-view convolutional neural networks for 3d shape recognition. [Paper]

• 2015-IROS - Voxnet: A 3d convolutional neural network for real-time object recognition. [Paper]

• 2015-IROS - VoxNet: A 3D convolutional neural network for realtime object recognition. [Paper]

• 2016-CVPR - Volumetric and multi-view cnns for object classification on 3d data. [Paper]

• 2016-ECCV - Vconv-DAE: Deep volumetric shape learning without object labels. [Paper]

• 2016-ECCV - Learning a predictable and generative vector representation for objects. [Paper]

• 2016-ECCV - 3d-r2n2: A unified approach for single and multi-view 3d object reconstruction. [Paper]

• 2016-NIPS - Learning a probabilistic latent space of object shapes via 3D generative adversarial modeling. [Paper][Code]

• 2017-BMVC - Orientation-boosted voxel nets for 3d object recognition. [Paper]

• 2020 - UCLID-Net: Single View Reconstruction in Object Space. [Paper]

• 2020-ECCV - CoReNet: Coherent 3D Scene Reconstruction from a Single RGB Image. [Paper]

• 2020-ACCV - DeepVoxels++: Enhancing the Fidelity of Novel View Synthesis from 3D Voxel Embeddings. [Paper]

Sparse Voxels

• 2016-NIPS - Fpnn: Field probing neural networks for 3d data. [Paper]

• 2017-3DV - Hierarchical surface prediction for 3d object reconstruction. [Paper]

• 2017-CVPR - Octnet: Learning deep 3d representations at high resolutions. [Paper]

• 2017-ICCV - Octree generating networks: Efficient convolutional architectures for high-resolution 3d outputs. [Paper]

• 2017-TOG - O-cnn: Octree-based convolutional neural networks for 3d shape analysis. [Paper]

• 2018-TOG - Adaptive O-CNN: A patch-based deep representation of 3D shapes. [Paper]

• 2020-Arxiv - Neural Sparse Voxel Fields. [Paper]

• 2020-TPAMI - Generative VoxelNet: Learning Energy-Based Models for 3D Shape Synthesis and Analysis. [Paper][Code]

Surface-based representation

Point-based

• 2017-CVPR - A point set generation network for 3d object reconstruction from a single image. [Paper]

• 2017-CVPR - Pointnet: Deep learning on point sets for 3d classification and segmentation. [Paper]

• 2017-ICCV - Escape from cells: Deep kd-networks for the recognition of 3D point cloud models. [Paper]

• 2017-NIPS - PointNet++: Deep hierarchical feature learning on point sets in a metric space. [Paper]

• 2018-CVPR - Foldingnet: Point cloud auto-encoder via deep grid deformation. [Paper]

• 2018-CVPR - Point cloud upsampling network. [Paper]

• 2018-CVPR - Sparse lattice networks for point cloud processing. [Paper]

• 2018-CVPR - Tangent convolutions for dense prediction in 3d. [Paper] [Tf-Code]

• 2018-ICML - Learning Representations and Generative Models For 3D Point Clouds. [Paper] [Tf-Code]

• 2018 - Point Cloud GAN. [Paper] [Py-Tf-Hybrid-Code]

• 2018-TOG - Point convolutional neural networks by extension operators. [Paper]

• 2018-EuroGraph - PCPNet: Learning Local Shape Properties from Raw Point Clouds. [Paper][Code]

• 2019-WACV - High Fidelity Semantic Shape Completion for Point Clouds using Latent Optimization. [Paper]

• 2019-CVPR - Patch-based progressive 3D point set upsampling. [Paper]

• 2019-CVPR - PU-GAN: a point cloud upsampling adversarial network. [Paper]

• 2019-CVPR - PointConv: Deep Convolutional Networks on 3D Point Clouds. [Paper][Code]

• 2019-CVPR - Supervised Fitting of Geometric Primitives to 3D Point Clouds. [Paper][Code]

• 2019-CVPR - PartNet: A Recursive Part Decomposition Network for Fine-grained and Hierarchical Shape Segmentation. [Paper][Code]

• 2019-CVPR - Modeling Local Geometric Structure of 3D Point Clouds using Geo-CNN. [Paper]

• 2019-CVPR - PointWeb: Enhancing Local Neighborhood Features for Point Cloud Processing. [Paper][Code]

• 2019-ICCV - DiscoNet: Shapes learning on disconnected manifolds for 3D editing. [Paper]

• 2019-ICCV - Interpolated Convolutional Networks for 3D Point Cloud Understanding. [Paper]

• 2019-NIPS - Point-Voxel CNN for Efficient 3D Deep Learning. [Paper]

• 2019-TOG - LOGAN: Unpaired Shape Transform in Latent Overcomplete Space. [Paper][Code]

• 2019 - ConvPoint: Continuous Convolutions for Point Cloud Processing. [Paper] [Py-Code]

• 2019-CGF - PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point Clouds. [Paper][Code]

• 2020-ICLR - Unpaired Point Cloud Completion on Real Scans using Adversarial Training. [Paper][Code]

• 2020-CVPR - AdaCoSeg: Adaptive Shape Co-Segmentation with Group Consistency Loss. [Paper]

• 2020-CVPR - PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection. [Paper][Code]

• 2020-ICRA - Dilated Point Convolutions: On the Receptive Field Size of Point Convolutions on 3D Point Clouds. [Paper]

• 2020-ECCV - SoftPoolNet: Shape Descriptor for Point Cloud Completion and Classification. [Paper]

• 2020-ECCV - Discrete Point Flow Networks for Efficient Point Cloud Generation. [Paper]

• 2020-ECCV - Learning Gradient Fields for Shape Generation. [Paper][Code]

• 2020-ECCV - Intrinsic Point Cloud Interpolation via Dual Latent Space Navigation. [Paper][Code]

• 2020-ECCV - Quaternion Equivariant Capsule Networks for 3D Point Clouds. [Paper]

• 2020-ECCV - SoftPoolNet: Shape Descriptor for Point Cloud Completion and Classification. [Paper]

• 2020-ECCV - DeepFit: 3D Surface Fitting via Neural Network Weighted Least Squares. [Paper]

• 2020-ECCV - Multimodal Shape Completion via Conditional Generative Adversarial Networks. [Paper]

• 2020-Arxiv - Point Cloud Completion by Skip-attention Network with Hierarchical Folding. [Paper]

• 2020-Arxiv - Detail Preserved Point Cloud Completion via Separated Feature Aggregation. [Paper]

• 2020-Arxiv - Point Cloud Completion by Learning Shape Priors. [Paper]

• 2020-Arxiv - KAPLAN: A 3D Point Descriptor for Shape Completion. [Paper]

• 2020-Arxiv - Self-Supervised Learning of Point Clouds via Orientation Estimation. [Paper]

• 2020-Arxiv - CaSPR: Learning Canonical Spatiotemporal Point Cloud Representations. Paper

• 2020-Arxiv - LPMNet: Latent Part Modification and Generation for 3D Point Clouds. [Paper]

• 2020-Arxiv - Weakly-supervised 3D Shape Completion in the Wild. [Papr]

• 2020-Arxiv - CAD-PU: A Curvature-Adaptive Deep Learning Solution for Point Set Upsampling. [Paper]

• 2020-Arxiv - Skeleton-bridged Point Completion: From Global Inference to Local Adjustment. [Paper]

• 2020-Arxiv - Hausdorff Point Convolution with Geometric Priors. [Paper]

Mesh-based

• 2016-ECCV - Deep learning 3D shape surfaces using geometry images. [Paper]

• 2016-NIPS - Learning shape correspondence with anisotropic convolutional neural networks. [Paper]

• 2017-TOG - Convolutional neural networks on surfaces via seamless toric covers. [Paper]

• 2017-CVPR - Geometric deep learning on graphs and manifolds using mixture model cnns. [Paper]

• 2017-ICCV - Directionally convolutional networks for 3D shape segmentation. [Paper]

• 2018-CVPR - Feastnet: Feature-steered graph convolutions for 3d shape analysis. [Paper]

• 2018-CVPR - SplineCNN: Fast geometric deep learning with continuous b-spline kernels. [Paper]

• 2018-CVPR - AtlasNet: A Papier-Mâché Approach to Learning 3D Surface Generation. [Paper]

• 2018-ECCV - Pixel2Mesh: Generating 3D mesh models from single RGB images. [Paper]

• 2018-Arxiv - Convolutional neural networks on 3d surfaces using parallel frames. [Paper]

• 2019-TOG - MeshCNN: a network with an edge. [Paper]

• 2019-ICCV - Pixel2Mesh++: Multi-view 3D mesh generation via deformation. [Paper]

• 2019-ICCVW - Learning to Reconstruct Symmetric Shapes using Planar Parameterization of 3D Surface. [Paper]

• 2020-ECCV - DR-KFS: A Differentiable Visual Similarity Metric for 3D Shape Reconstruction. [Paper]

• 2020-ECCV - Weakly-supervised 3D Shape Completion in the Wild. [Paper]

Implicit representation

• 2019-CVPR - Learning Implicit Fields for Generative Shape Modeling. [Paper] [Code]

• 2019-CVPR - Occupancy networks: Learning 3D reconstruction in function space. [Paper] [Code]

• 2019-CVPR - DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation. [Paper] [Code]

• 2019-ICCV - PIFu: Pixel-Aligned Implicit Function for High-Resolution Clothed Human Digitization. [Paper]

• 2019-ICCV - Learning Shape Templates with Structured Implicit Functions. [Paper]

• 2019-NIPS - Scene Representation Networks: Continuous 3D-Structure-Aware Neural Scene Representations. [Paper] (implicity not verified)

• 2019-NIPS - DISN: Deep implicit surface network for high-quality single-view 3D reconstruction. [Paper] [Code]

• 2019-NIPS - Learning to infer implicit surfaces without 3D supervision. [Paper]

• 2019-Arxiv - Deep structured implicit functions. [Paper]

• 2019-Arxiv - Learning shape templates with structured implicit functions. [Paper]

• 2020-Arxiv - Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction. [Paper]

• 2020 - MeshSDF: Differentiable Iso-Surface Extraction. [Paper]

• 2020-ECCV - Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction. [Paper][Code]

• 2020-ECCV - Ladybird: Quasi-Monte Carlo Sampling for Deep Implicit Field Based 3D Reconstruction with Symmetry. [Paper]

• 2020-ECCV - Pix2Surf: Learning Parametric 3D Surface Models of Objects from Images. [Paper]

• 2020-ECCV - Points2Surf: Learning Implicit Surfaces from Point Cloud Patches. [Paper][Code]

• 2020-ECCV - Curriculum DeepSDF. [Paper][Code]

• 2020-NIPS - Neural Unsigned Distance Fields for Implicit Function Learning. [Paper]

• 2020-NIPS - Learning Implicit Functions for Topology-Varying Dense 3D Shape Correspondence. [Paper]

• 2020 - PatchNets: Patch-Based Generalizable Deep Implicit 3D Shape Representations. [Paper]

• 2020-Arxiv - SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces from RGB Images. [Paper]

• 2020-Arxiv - Implicit Feature Networks for Texture Completion from Partial 3D Data. [Paper]

• 2020-Arxiv - Overfit Neural Networks as a Compact Shape Representation. [Paper]

• 2020-MM - Vaccine-style-net: Point Cloud Completion in Implici Continuous Function Space. [Paper]

• 2020-Arxiv - Learning Occupancy Function from Point Clouds for Surface Reconstruction. [Paper]

• 2020-Arxiv - DUDE: Deep Unsigned Distance Embeddings for Hi-Fidelity Representation of Complex 3D Surfaces. [Paper]

• 2020-Arxiv - Deep Implicit Templates for 3D Shape Representation. [Paper]

• 2020-Arxiv - NeuralFusion: Online Depth Fusion in Latent Space. [Paper]

• 2021-Arxiv - Neural Geometric Level of Detail: Real-time Rendering with Implicit 3D Shapes. [Paper][Code]

• 2021-Arxiv - Secrets of 3D Implicit Object Shape Reconstruction in the Wild. [Paper]

• 2021-Arxiv - Generative Models as Distributions of Functions. [paper]

• 2021-Arxiv - Shelf-Supervised Mesh Prediction in the Wild. [Paper]

Structure/Part-based representation

• 2011-EG - Symmetry hierarchy of man-made objects. [Paper]

• 2014-TOG - Structure-aware shape processing. [Paper]

• 2015-CGF - Analysis and synthesis of 3d shape families via deep-learned generative models of surfaces. [Paper]

• 2017-CGF - The shape variational autoencoder: A deep generative model of part-segmented 3d objects. [Paper]

• 2018-TOG - Global-to-local generative model for 3d shapes. [Paper][Tf-Code]

• 2019-ICCV - Composite Shape Modeling via Latent Space Factorization. [Paper]

• 2019 - Learning structural graph layouts and 3d shapes for long span bridges 3d reconstruction. [Paper]

• 2019-ICCV - CompoNet: Learning to Generate the Unseen by Part Synthesis and Composition. [Paper][Tf-Code]

• 2019-ICCV - BAE-NET: Branched Autoencoder for Shape Co-Segmentation. [Paper][Tf-Code]

• 2019-TOG - Structurenet: Hierarchical graph networks for 3d shape generation. [Paper] [Tf-Code]

• 2019-TOG - SAGNet: Structure-aware Generative Network for 3D-Shape Modeling. [Paper] [Tf-Code]

• 2019-TOG - SDM-NET: Deep generative network for structured deformable mesh. [Paper] [Code]

• 2020-CVPR - BSP-Net: Generating compact meshes via binary space partitioning. [Paper][Code]

• 2020-CVPR - PQ-NET: A Generative Part Seq2Seq Network for 3D Shapes. [Paper][Code]

• 2020-CVPR - SSRNet: Scalable 3D Surface Reconstruction Network. Paper

• 2020-CVPR - Neural Implicit Embedding for Point Cloud Analysis. Paper

• 2020-CVPR - FroDO: From Detections to 3D Objects. Paper

• 2020-CVPR - StructEdit: Learning Structural Shape Variations. [Paper][Code]

• 2020-SigAsia - ShapeAssembly: Learning to Generate Programs for 3D Shape Structure Synthesis. [Paper][Code]

• 2020-Arxiv - A Simple and Scalable Shape Representation for 3D Reconstruction. [Paper]

• 2020-Arxiv - Topology-Aware Single-Image 3D Shape Reconstruction. [Paper]

• 2020-Arxiv - Info3D: Representation Learning on 3D Objects using Mutual Information Maximization and Contrastive Learning. Paper

• 2020-Arxiv - COALESCE: Component Assembly by Learning to Synthesize Connections. Paper

• 2020-Arxiv - DSM-Net: Disentangled Structured Mesh Net for Controllable Generation of Fine Geometry. [Paper]

• 2020-ECCV - GSIR: Generalizable 3D Shape Interpretation and Reconstruction. [Paper]

• 2020-ECCV - Learning 3D Part Assembly from a Single Image. [Paper][Code]

• 2020-Arxiv - TM-NET: Deep Generative Networks for Textured Meshes. [Paper]

• 2021-Arxiv - Towards Generalising Neural Implicit Representations. [Paper]

• 2021- Extending StructureNet to Generate Physically Feasible 3D Shapes. [Paper]

• 2021- How to represent part-whole hierarchies in a neural network. [Paper]

Deformation-based methods

• 2016-TOG - Efficient and flexible deformation representation for data-driven surface modeling. [Paper]

• 2018-CVPR - Variational autoencoders for deforming 3d mesh models. [Paper]

• 2019-TVCG - Sparse data driven mesh deformation. [Paper]

• 2018-AAAI - Mesh-based autoencoders for localized deformation component analysis. [Paper]

• 2018-TOG - Automatic unpaired shape deformation transfer. [Paper]

• 2019-Arxiv - NASA: Neural articulated shape approximation. [Paper]

• 2020-ECCV - LIMP: Learning Latent Shape Representations with Metric Preservation Priors. [Paper]

• 2020-Siggraph - Point2Mesh: A Self-Prior for Deformable Meshes. [Paper][Code]

• 2020-Arxiv - Better Patch Stitching for Parametric Surface Reconstruction. [Paper]

• 2020-NIPS - ShapeFlow: Learnable Deformations Among 3D Shapes. [Paper]

• 2020-ECCV - NASA Neural Articulated Shape Approximation. [Paper]

Primitve-based methods

• 1971-CSSC - Visual perception by computer.

• 1986-TOG - Constructive solid geometry for polyhedral objects. [Paper]

• 2016 - Volumetric Hierarchical Approximate Convex Decomposition. [Paper]

• 2017-CVPR - Learning shape abstractions by assembling volumetric primitives. [Paper][Code][Py-Code]

• 2017-ICCV - 3dprnn: Generating shape primitives with recurrent neural networks. [Paper][Code]

• 2017-TOG - GRASS: Generative recursive autoencoders for shape structures. [Paper][Code][Py-Code]

• 2017-TOG - ComplementMe: Weakly-Supervised Component Suggestions for 3D Modeling. [Paper][Code]

• 2018-CVPR - Csgnet: Neural shape parser for constructive solid geometry. [Paper][Code]

• 2018-CVPR - Im2struct: Recovering 3d shape structure from a single RGB image. [Paper][Code]

• 2018-ECCV - Physical Primitive Decomposition. [Paper]

• 2018-NIPS - Learning to Infer Graphics Programs from Hand-Drawn Images. [Paper][Code]

• 2018-SigAsia - FrankenGAN: Guided Detail Synthesis for Building Mass Models Using Style-Synchonized GANs. [Paper][Code]

• 2018-TOG - InverseCSG: Automatic Conversion of 3D Models to CSG Trees. [Paper]

• 2018-3DV - Parsing Geometry Using Structure-Aware Shape Templates. [Paper][Code]

• 2019-CVPR - Supervised fitting of geometric primitives to 3d point clouds. [Paper][Code]

• 2019-CVPR - Superquadrics Revisited: Learning 3D Shape Parsing Beyond Cuboids. [Paper][Code]

• 2019-CVPR - Unsupervised Primitive Discovery for Improved 3D Generative Modeling. [Paper]

• 2019-ICCV - Learning Shape Templates with Structured Implicit Functions. [Paper][Code]

• 2019-NIPS - Learning elementary structures for 3d shape generation and matching. [Paper][Code]

• 2019-ICLR - Learning to Infer and Execute 3D Shape Programs. [Paper][Code]

• 2019-IJCV - Learning Single-Image 3D Reconstruction by Generative Modelling of Shape, Pose and Shading. [Paper][Code]

• 2019-IGARSS - Primitive-Based 3D Building Modeling, Sensor Simulation, and Estimation. [Paper]

• 2019-TOG - Learning Adaptive Hierarchical Cuboid Abstractions of 3D Shape Collections. [Paper][Code]

• 2019-TPAMI - PRS-Net: Planar Reflective Symmetry Detection Net for 3D Models. [Paper]

• 2020-CVPR - Learning Unsupervised Hierarchical Part Decomposition of 3D Objects from a Single RGB Image. [Paper][Code]

• 2020-CVPR - DualSDF: Semantic Shape Manipulation using a Two-Level Representation. [Paper][Code]

• 2020-CVPR - CvxNet: Learnable Convex Decomposition. [Paper][Code]

• 2020-CVPR - Local Deep Implicit Functions for 3D Shape. [Paper][Code]

• 2020-CVPR - Deep Parametric Shape Predictions using Distance Fields. [Paper][Code]

• 2020-CVPR - Learning Generative Models of Shape Handles. [Paper]

• 2020-ECCV - ParSeNet: A Parametric Surface Fitting Network for 3D Point Clouds. [Paper][Code]

• 2020-ECCV - Label-Efficient Learning on Point Clouds using Approximate Convex Decompositions. [Paper][Code]

• 2020-BMVC - 3D-GMNet: Single-View 3D Shape Recovery as A Gaussian Mixture. [Paper]

• 2020-IWOBI - Recovery of Superquadrics from Range Images using Deep Learning: A Preliminary Study. [Paper]

• 2020-NIPS - Neural Star Domain as Primitive Representation. [Paper]

• 2020-NIPS - PIE-NET: Parametric Inference of Point Cloud Edges. [Paper][Code]

• 2020-NIPS - UCSG-NET - Unsupervised Discovering of Constructive Solid Geometry Tree. [Paper][Code]

• 2020-CGF - Learning Generative Models of 3D Structures. [Paper]

• 2020 - Geometric Primitives in LiDAR Point Clouds: A Review. [Paper]

• 2020 - Unsupervised Deep Learning for Primitive-Based Shape Abstraction. [Paper]

• 2020 - Dynamic Plane Convolutional Occupancy Networks. [Paper]

• 2020-Arxiv - Learning to Infer Shape Programs Using Latent Execution Self Training. [Paper]

• 2020-IEEEAccess - Learning to predict superquadric parameters from depth images with explicit and implicit supervision. [Paper]

• 2021 - Facilitated editing of generative design geometry in computer aided design user interface. [Paper]

• 2021 - Mixture of Volumetric Primitives for Efficient Neural Rendering. [Paper]

• 2021-Arxiv - EllipsoidNet: Ellipsoid Representation for Point Cloud Classification and Segmentation. [Paper]

• 2021- A deep neural network based shape unification to define a 3-Dimensional shape. [Paper]

Survey Papers

• 2019-Arxiv - Deep learning for 3D point clouds: A survey. [Paper]

• 2020-Arxiv - A Survey on Deep Geometry Learning: From a Representation Perspective. [Paper]

• 2020-Arxiv - Image-based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era. [Paper]

• 2020-MTA - Single image 3D object reconstruction based on deep learning: A review. [Paper]

• 2020-CG - Single-View 3D reconstruction: A Survey of deep learning methods. [Paper]

• 2021-Arxiv - Attention Models for Point Clouds in Deep Learning: A Survey. [Paper]

• 2021-MTA - A survey of recent 3D scene analysis and processing methods. [Paper]

• 2021-Arxiv - A comprehensive survey on point cloud registration. [Paper]

Misc

• Learning to Generate 3D Training Data - [Thesis]

• 2020-Arxiv - Deep Optimized Priors for 3D Shape Modeling and Reconstruction. [Paper]

Datasets

Part Labels

• 2009-TOG - A Benchmark for 3D Mesh Segmentation. [Paper][Dataset]

• 2010-TOG - Learning 3D Mesh Segmentation and Labeling. [Paper][Dataset]

• 2012-TOG - Active Co-Analysis of a Set of Shapes. [Paper][Dataset][Project Page]

• 2016-TOG - Point labels on ShapeNet Data - A Scalable Active Framework for Region Annotation in 3D Shape Collections. [Paper][Dataset]

• 2017-CVPR - Mesh labels on ShapeNet Data - 3D Shape Segmentation with Projective Convolutional Networks. [Paper][Dataset][Code]

• 2019-CVPR - PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understanding. [Paper][Dataset]

Other Resources

• awesome point cloud analysis