VConv-DAE: Deep Volumetric Shape Learning Without Object Labels

By Abhishek Sharma, Oliver Grau, Mario Fritz. 2016

297 citations recorded by the source corpus.

With the advent of affordable depth sensors, 3D capture becomes more and more ubiquitous and already has made its way into commercial products. Yet, capturing the geometry or complete shapes of everyday objects using scanning devices (e.g. Kinect) still comes with several challenges that result in noise or even incomplete shapes.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Learning Descriptor Networks for 3D Shape Synthesis and Analysis (2018, methodology): “e 4: 3D object classiﬁcation on ModelNet10 dataset Method Accuracy Geometry Image [23] 88.4% PANORAMA-NN [19] 91.1% ECC [22] 90.0% 3D ShapeNets [29] 83.5% DeepPano [21] 85.5% SPH [10] 79.8% VConv-DAE [20] 80.5% 3D-GAN [28] 91.0% 3D DescriptorNet (ours) 92.4% 5. Conclusion We propose the 3D Descripto”
Multi-View Silhouette and Depth Decomposition for High Resolution 3D Object Representation (2018, background): “Deep generative models have recently achieved strong performance in hallucinating diverse 3D object shapes, capturing their overall structure and rough texture [3, 37, 46].”
Octree Generating Networks: Efficient Convolutional Architectures for High-resolution 3D Outputs (2017, methodology): “Applications include single- and multi-view 3D object reconstruction trained in supervised [14, 5, 7, 17] and unsupervised [37, 12, 19] ways, probabilistic generative modeling of 3D shapes [36, 35, 29], semantic segmentation [2, 4] and shape deformation [39].”
Shape Completion Using 3D-Encoder-Predictor CNNs and Shape Synthesis (2016, background): “Since then different versions of 3D-CNN architectures have been proposed in order to improve classiﬁcation accuracy [22, 31, 34], obtain object proposals [37], or match local 3D geometry [46].”
Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (2016, methodology): “Our representation outperforms other features learned without supervision by a large margin (83.3% vs. 75.5% on ModelNet40, and 91.0% vs 80.5% on ModelNet10) [Girdhar et al., 2016, Sharma et al., 2016].”
Deformable Shape Completion with Graph Convolutional Autoencoders (2017, methodology): “While several methods based on (volumetric) convolutional neural networks have been proposed for completing man-made rigid objects (see [12, 51, 55, 61, 49]), they struggle at handling deformable shapes.”
SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (2017, background): “ion for 3D objects using images and CAD objects which are used for generating 3D shapes from an image. A volumetric denoising auto-encoder is demonstrated for 3D shape completion from noisy inputs in [26]. Choy et al. propose a 3D recurrent reconstruction neural network for 3D shape creation from si”
3D2SeqViews: Aggregating Sequential Views for 3D Global Feature Learning by CNN With Hierarchical Attention Aggregation (2019, methodology): “Sharma et al. [17] employed fully convolutional denoising autoencoder to robustly perform unsupervised global feature learning by decomposing and reconstructing voxelized 3D shapes.”
FoldingNet: Interpretable Unsupervised Learning on 3D Point Clouds (2017, result): “The results of [8,17, 25,43] are according to the report in [1, 53].”
Global-Local Bidirectional Reasoning for Unsupervised Representation Learning of 3D Point Clouds (2020, unknown): “40 VConv-DAE [39] voxel 75.”
PCN: Point Completion Network (2018, methodology): “While most existing learning-based methods [9, 16, 43, 46, 49, 53, 55, 58] represents shapes using voxels, which are convenient for convolutional neural networks, our method uses point clouds, which preserve complete geometric information about the shapes while being memory efficient.”
Improved Adversarial Systems for 3D Object Generation and Reconstruction (2017, background): “ed for the task. Recently, techniques for unsupervised learning of deep networks trained on large datasets of synthetic 3D shapes have produced models capable of representing the detailed shape space [1,2]. This allows the generation of strikingly realistic sample shapes for simulation or visualizat”
Multiresolution Tree Networks for 3D Point Cloud Processing (2018, unknown): “…91.2 MRTNet (4K pts) 91.7 KDNet (32K pts) [24] 91.8 Point-based methods (with normals) PointNet++ (5K pts) [32] 91.9 Voxel-based methods OctNet [33] 86.5 O-CNN [42] 90 75.5 T-L Network[14] 74.4 VConv-DAE [34] 75.5 3D-GAN [43] 83.3 MRTNet-VAE (Ours) 86.4 (c) Unsupervised representation learning .”
Learning a Hierarchical Latent-Variable Model of 3D Shapes (2017, unknown): “4% VConv-DAE [37] 80.”
Self-Supervised Deep Learning on Point Clouds by Reconstructing Space (2019, unknown): “Besides Autoencoders on voxelized representations [29] in which conventional loss functions can be applied per-voxel, Autoencoders have also been applied on raw point clouds [37, 15].”