pith. sign in

arxiv: 1907.05091 · v1 · pith:EZAB7OJ5new · submitted 2019-07-11 · 💻 cs.CV

Efficient Semantic Scene Completion Network with Spatial Group Convolution

Jiahui Zhang , Hao Zhao , Anbang Yao , Yurong Chen , Li Zhang , Hongen Liao This is my paper

Pith reviewed 2026-05-24 23:17 UTC · model grok-4.3

classification 💻 cs.CV
keywords semantic scene completionspatial group convolution3D sparse convolutionSUNCG datasetefficient 3D networksvoxel groupingmultiscale architecturecoarse-to-fine prediction
0
0 comments X

The pith

Spatial Group Convolution accelerates 3D semantic scene completion by partitioning voxels into groups for independent sparse convolutions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Spatial Group Convolution to speed up 3D dense prediction tasks by splitting voxels into spatial groups and running sparse convolution separately on each. This reduces overall computation because only valid voxels within groups are processed. The operation is applied to semantic scene completion, which predicts a full labeled 3D volume from one depth image. A multiscale sparse convolutional network using coarse-to-fine prediction is built around SGC. On the SUNCG dataset the resulting system reaches state-of-the-art accuracy at high speed.

Core claim

Spatial Group Convolution partitions the input voxels into different spatial groups and performs 3D sparse convolution independently on each group. When embedded in a multiscale architecture that employs a coarse-to-fine prediction strategy, the resulting network predicts complete semantic 3D scenes from single depth images while delivering state-of-the-art performance and fast speed on the SUNCG dataset.

What carries the argument

Spatial Group Convolution (SGC), which divides voxels into spatial groups and applies 3D sparse convolution only within each group to cut computation.

If this is right

  • Computation drops substantially because convolution is restricted to valid voxels inside each separate group.
  • State-of-the-art accuracy is reached on the SUNCG benchmark for semantic scene completion.
  • Inference runs at high speed suitable for practical deployment.
  • SGC operates orthogonally to channel-wise group convolution and can be combined with it.
  • The multiscale coarse-to-fine design further improves both efficiency and final label quality.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • SGC could transfer to other voxel-grid tasks such as 3D object detection with only minor adaptation.
  • Dynamic, content-dependent grouping might shrink the accuracy penalty further.
  • The spatial partitioning idea suggests similar efficiency gains are possible in 2D dense prediction by grouping pixels.
  • Hardware support for sparse group-wise operations would compound the reported speed advantage.

Load-bearing premise

That partitioning voxels into spatial groups produces only a slight accuracy drop while delivering large compute savings, without the grouping choice itself requiring task-specific tuning that offsets the reported gains.

What would settle it

Measuring accuracy and runtime on SUNCG when the number of spatial groups is varied; if accuracy falls sharply for the group counts that produce the claimed speedups, the central claim does not hold.

Figures

Figures reproduced from arXiv: 1907.05091 by Anbang Yao, Hao Zhao, Hongen Liao, Jiahui Zhang, Li Zhang, Yurong Chen.

Figure 1
Figure 1. Figure 1: A 3D scene image from the SUNCG dataset. Left is the ground truth image. Right is a sampled image with only 30% voxels reserved. Giving only partial voxels does not prevent humans in reasoning the overall semantic information, but it imposes a challenge to recognize small objects such as chair’s leg. (Best viewed in color) layer and Abstracting Module are designed to generate voxels which are absent in inp… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of SGC. Feature maps are partitioned uniformly into different groups along the spatial dimensions (only two groups are shown here). 3D CNNs are conducted on different groups and give the final dense prediction for all voxels. Weights are shared between different groups. In the implementation of SGC, we partition features along the spatial dimen￾sions and then stack different groups along the b… view at source ↗
Figure 3
Figure 3. Figure 3: Network architecture for semantic scene completion. Taking flipped TSDF as input, the network predicts occupancy and object labels in 1/4 size. The resolution of each layer is marked nearby. Parameters of each layer are shown in the order of (filter size, stride, output channel). Dense deconvolution layers can generate new voxels. The Abstracting module can abstract non-trivial voxels to high resolution ac… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results of our network and SSCNet. We achieve obviously much better results, such as predictions around object boundaries [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Histograms of learned weight values of SCN and SGC with different groups. The first row shows the statistics of the first convolution layer, and the second row shows that of the last convolution layer. Filters of SGC have “sharper” histograms. 5 Discussion 5.1 What does Spatial Group Convolution learn? In [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of SGC with fixed pattern partition. (a) shows that for a 3 × 3 kernel, an “X” shape filter is learned when partitioning voxels into two groups. (b) shows the learned 3 × 3 × 3 filters in [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
read the original abstract

We introduce Spatial Group Convolution (SGC) for accelerating the computation of 3D dense prediction tasks. SGC is orthogonal to group convolution, which works on spatial dimensions rather than feature channel dimension. It divides input voxels into different groups, then conducts 3D sparse convolution on these separated groups. As only valid voxels are considered when performing convolution, computation can be significantly reduced with a slight loss of accuracy. The proposed operations are validated on semantic scene completion task, which aims to predict a complete 3D volume with semantic labels from a single depth image. With SGC, we further present an efficient 3D sparse convolutional network, which harnesses a multiscale architecture and a coarse-to-fine prediction strategy. Evaluations are conducted on the SUNCG dataset, achieving state-of-the-art performance and fast speed. Code is available at https://github.com/zjhthu/SGC-Release.git

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces Spatial Group Convolution (SGC), an operation that partitions input voxels into spatial groups and applies independent 3D sparse convolutions within each group to reduce computation for dense 3D prediction. It integrates SGC into a multiscale 3D sparse convolutional network using a coarse-to-fine strategy for the semantic scene completion task and reports state-of-the-art results with fast inference on the SUNCG dataset. The code is released publicly.

Significance. If the efficiency gains hold with only minor accuracy degradation and without hidden per-task tuning costs for group formation, SGC would be a practical, orthogonal acceleration technique for 3D sparse convolutions. Public code release is a clear strength that supports verification and extension.

major comments (2)
  1. [SGC definition and algorithm] The central efficiency claim rests on the spatial partitioning step, yet the manuscript provides no explicit description (in the method section or algorithm) of whether groups are formed via fixed grid, occupancy statistics, or learned parameters; without this, it is impossible to assess whether group selection itself incurs search or validation cost comparable to the reported savings.
  2. [Experiments and results] The claim of 'slight loss of accuracy' is load-bearing for the contribution, but the experiments section lacks a direct ablation isolating the accuracy-runtime trade-off of SGC versus standard sparse convolution on the same backbone; quantitative deltas (e.g., IoU drop and FLOPs reduction on SUNCG) are required to substantiate the claim.
minor comments (1)
  1. [Implementation details] Notation for group count and group size is introduced without a clear table or equation reference; adding a short summary table of hyper-parameters used on SUNCG would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments below and will incorporate clarifications and additional experiments into a revised manuscript.

read point-by-point responses

  1. Referee: [SGC definition and algorithm] The central efficiency claim rests on the spatial partitioning step, yet the manuscript provides no explicit description (in the method section or algorithm) of whether groups are formed via fixed grid, occupancy statistics, or learned parameters; without this, it is impossible to assess whether group selection itself incurs search or validation cost comparable to the reported savings.

    Authors: We agree that an explicit description of group formation is needed for reproducibility and to confirm zero overhead. SGC performs a deterministic fixed-grid partitioning of the 3D volume into non-overlapping spatial blocks before applying independent sparse convolutions; no occupancy statistics or learned parameters are used for group assignment. We will add a precise textual description, diagram, and algorithm box in the revised Method section to document this process and its O(1) cost. revision: yes

  2. Referee: [Experiments and results] The claim of 'slight loss of accuracy' is load-bearing for the contribution, but the experiments section lacks a direct ablation isolating the accuracy-runtime trade-off of SGC versus standard sparse convolution on the same backbone; quantitative deltas (e.g., IoU drop and FLOPs reduction on SUNCG) are required to substantiate the claim.

    Authors: We acknowledge that a controlled head-to-head ablation on the identical backbone is required. In the revision we will add a dedicated table and paragraph reporting mIoU, IoU, and FLOPs (or equivalent runtime) for the baseline sparse-convolution network versus the SGC variant on SUNCG, thereby quantifying the accuracy-runtime trade-off directly. revision: yes

Circularity Check

0 steps flagged

No circularity; engineering proposal validated on external benchmark

full rationale

The paper introduces Spatial Group Convolution (SGC) as a practical optimization that partitions voxels and applies independent sparse convolutions within groups. No equations, fitted parameters, or predictions are defined in terms of themselves. The central claim (efficiency with minor accuracy loss on semantic scene completion) is supported by empirical evaluation on the SUNCG dataset rather than any self-referential derivation or self-citation chain. The method is presented as an orthogonal engineering technique, not a mathematical result derived from prior fitted quantities or uniqueness theorems.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented physical entities are stated. The new operator itself is the contribution.

pith-pipeline@v0.9.0 · 5691 in / 1051 out tokens · 14497 ms · 2026-05-24T23:17:28.451880+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages · 5 internal anchors

  1. [1]

    3D Point Cloud Classification and Segmentation using 3D Modified Fisher Vector Representation for Convolutional Neural Networks

    Ben-Shabat, Y., Lindenbaum, M., Fischer, A.: 3d point cloud classification and seg- mentation using 3d modified fisher vector representation for convolutional neural networks. arXiv preprint arXiv:1711.08241 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  2. [2]

    In: 2017 International Conference on 3D Vision (3DV)

    Chang, A., Dai, A., Funkhouser, T., Halber, M., Niebner, M., Savva, M., Song, S., Zeng, A., Zhang, Y.: Matterport3d: Learning from rgb-d data in indoor environ- ments. In: 2017 International Conference on 3D Vision (3DV). pp. 667–676. IEEE (2017)

    work page 2017

  3. [3]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Chollet, F.: Xception: Deep learning with depthwise separable convolutions. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 1251–1258 (2017)

    work page 2017

  4. [4]

    In: Proc

    Dai, A., Qi, C.R., Nießner, M.: Shape completion using 3d-encoder-predictor cnns and shape synthesis. In: Proc. IEEE Conf. on Computer Vision and Pattern Recog- nition (CVPR). vol. 3 (2017)

    work page 2017

  5. [5]

    In: CVPR

    Dai, A., Ritchie, D., Bokeloh, M., Reed, S., Sturm, J., Nießner, M.: Scancomplete: Large-scale scene completion and semantic segmentation for 3d scans. In: CVPR. vol. 1, p. 2 (2018)

    work page 2018

  6. [6]

    In: Robotics and Automation (ICRA), 2017 IEEE International Conference on

    Engelcke, M., Rao, D., Wang, D.Z., Tong, C.H., Posner, I.: Vote3deep: Fast ob- ject detection in 3d point clouds using efficient convolutional neural networks. In: Robotics and Automation (ICRA), 2017 IEEE International Conference on. pp. 1355–1361. IEEE (2017)

    work page 2017

  7. [7]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Firman, M., Mac Aodha, O., Julier, S., Brostow, G.J.: Structured prediction of un- observed voxels from a single depth image. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 5431–5440 (2016)

    work page 2016

  8. [8]

    Compressing Deep Convolutional Networks using Vector Quantization

    Gong, Y., Liu, L., Yang, M., Bourdev, L.: Compressing deep convolutional networks using vector quantization. arXiv preprint arXiv:1412.6115 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  9. [9]

    1–11 (2015)

    Graham, B.: Sparse 3D convolutional neural networks pp. 1–11 (2015). https://doi.org/10.1109/TPAMI.2012.59

    work page doi:10.1109/tpami.2012.59 2015

  10. [10]

    CVPR (2018)

    Graham, B., Engelcke, M., van der Maaten, L.: 3d semantic segmentation with submanifold sparse convolutional networks. CVPR (2018)

    work page 2018

  11. [11]

    Submanifold Sparse Convolutional Networks

    Graham, B., van der Maaten, L.: Submanifold sparse convolutional networks. arXiv preprint arXiv:1706.01307 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  12. [12]

    Predicting Complete 3D Models of Indoor Scenes

    Guo, R., Zou, C., Hoiem, D.: Predicting complete 3d models of indoor scenes. arXiv preprint arXiv:1504.02437 (2015)

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  13. [13]

    arXiv preprint arXiv:1801.10585 (2018)

    Hackel, T., Usvyatsov, M., Galliani, S., Wegner, J.D., Schindler, K.: Inference, learning and attention mechanisms that exploit and preserve sparsity in convolu- tional networks. arXiv preprint arXiv:1801.10585 (2018)

    work page arXiv 2018

  14. [14]

    In: Advances in neural information processing systems

    Han, S., Pool, J., Tran, J., Dally, W.: Learning both weights and connections for efficient neural network. In: Advances in neural information processing systems. pp. 1135–1143 (2015)

    work page 2015

  15. [15]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition

    Han, X., Li, Z., Huang, H., Kalogerakis, E., Yu, Y.: High-resolution shape comple- tion using deep neural networks for global structure and local geometry inference. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition. pp. 85–93 (2017)

    work page 2017

  16. [16]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Handa, A., Patraucean, V., Badrinarayanan, V., Stent, S., Cipolla, R.: Under- standing real world indoor scenes with synthetic data. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 4077–4085 (2016)

    work page 2016

  17. [17]

    In: Proceedings of the International Conference on 3D Vision (2017) 16 Jiahui Zhang, Hao Zhao and et al

    H¨ ane, C., Tulsiani, S., Malik, J.: Hierarchical surface prediction for 3d object re- construction. In: Proceedings of the International Conference on 3D Vision (2017) 16 Jiahui Zhang, Hao Zhao and et al

    work page 2017

  18. [18]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    work page 2016

  19. [19]

    In: European Conference on Computer Vision

    He, K., Zhang, X., Ren, S., Sun, J.: Identity mappings in deep residual networks. In: European Conference on Computer Vision. pp. 630–645. Springer (2016)

    work page 2016

  20. [20]

    ArXiv e-prints (Apr 2017)

    Howard, A.G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., An- dreetto, M., Adam, H.: MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. ArXiv e-prints (Apr 2017)

    work page 2017

  21. [21]

    In: International conference on machine learning

    Ioffe, S., Szegedy, C.: Batch normalization: Accelerating deep network training by reducing internal covariate shift. In: International conference on machine learning. pp. 448–456 (2015)

    work page 2015

  22. [22]

    In: ICCV Workshops (2017)

    Johnston, A., Garg, R., Carneiro, G., Reid, I., vd Hengel, A.: Scaling cnns for high resolution volumetric reconstruction from a single image. In: ICCV Workshops (2017)

    work page 2017

  23. [23]

    In: 2017 IEEE International Conference on Computer Vision (ICCV)

    Klokov, R., Lempitsky, V.: Escape from cells: Deep kd-networks for the recognition of 3d point cloud models. In: 2017 IEEE International Conference on Computer Vision (ICCV). pp. 863–872. IEEE (2017)

    work page 2017

  24. [24]

    In: Advances in neural information processing systems

    Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep con- volutional neural networks. In: Advances in neural information processing systems. pp. 1097–1105 (2012)

    work page 2012

  25. [25]

    In: Proceedings of the 2Nd International Conference on Neural Information Processing Systems

    Le Cun, Y., Denker, J.S., Solla, S.A.: Optimal brain damage. In: Proceedings of the 2Nd International Conference on Neural Information Processing Systems. pp. 598–605. NIPS’89, MIT Press, Cambridge, MA, USA (1989)

    work page 1989

  26. [26]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Li, X., Liu, Z., Luo, P., Change Loy, C., Tang, X.: Not all pixels are equal: Difficulty- aware semantic segmentation via deep layer cascade. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 3193–3202 (2017)

    work page 2017

  27. [27]

    ArXiv e-prints (Jan 2018)

    Li, Y., Bu, R., Sun, M., Chen, B.: PointCNN. ArXiv e-prints (Jan 2018)

    work page 2018

  28. [28]

    In: Advances in Neural Information Processing Systems

    Li, Y., Pirk, S., Su, H., Qi, C.R., Guibas, L.J.: Fpnn: Field probing neural networks for 3d data. In: Advances in Neural Information Processing Systems. pp. 307–315 (2016)

    work page 2016

  29. [29]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Liu, F., Li, S., Zhang, L., Zhou, C., Ye, R., Wang, Y., Lu, J.: 3dcnn-dqn-rnn: A deep reinforcement learning framework for semantic parsing of large-scale 3d point clouds. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 5678–5687 (2017)

    work page 2017

  30. [30]

    In: Intelligent Robots and Systems (IROS), 2015 IEEE/RSJ International Conference on

    Maturana, D., Scherer, S.: Voxnet: A 3d convolutional neural network for real-time object recognition. In: Intelligent Robots and Systems (IROS), 2015 IEEE/RSJ International Conference on. pp. 922–928. IEEE (2015)

    work page 2015

  31. [31]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Qi, C.R., Su, H., Mo, K., Guibas, L.J.: Pointnet: Deep learning on point sets for 3d classification and segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 652–660 (2017)

    work page 2017

  32. [32]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Qi, C.R., Su, H., Nießner, M., Dai, A., Yan, M., Guibas, L.J.: Volumetric and multi-view cnns for object classification on 3d data. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 5648–5656 (2016)

    work page 2016

  33. [33]

    In: Advances in Neural Information Processing Systems

    Qi, C.R., Yi, L., Su, H., Guibas, L.J.: Pointnet++: Deep hierarchical feature learn- ing on point sets in a metric space. In: Advances in Neural Information Processing Systems. pp. 5105–5114 (2017)

    work page 2017

  34. [34]

    In: Proceedings of the International Conference on Computer Vision (2017)

    Qi, X., Liao, R., Jia, J., Fidler, S., Urtasun, R.: 3D Graph Neural Networks for RGBD Semantic Segmentation. In: Proceedings of the International Conference on Computer Vision (2017). https://doi.org/10.1109/ICCV.2017.556 Semantic Scene Completion with Spatial Group Convolution 17

    work page doi:10.1109/iccv.2017.556 2017

  35. [35]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Ren, M., Pokrovsky, A., Yang, B., Urtasun, R.: Sbnet: Sparse blocks network for fast inference. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 8711–8720 (2018)

    work page 2018

  36. [36]

    In: Proceedings of the International Conference on 3D Vision (2017)

    Riegler, G., Ulusoy, A.O., Bischof, H., Geiger, A.: Octnetfusion: Learning depth fusion from data. In: Proceedings of the International Conference on 3D Vision (2017)

    work page 2017

  37. [37]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Riegler, G., Ulusoy, A.O., Geiger, A.: Octnet: Learning deep 3d representations at high resolutions. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. vol. 3 (2017)

    work page 2017

  38. [38]

    In: International Conference on Medical image computing and computer-assisted intervention

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedi- cal image segmentation. In: International Conference on Medical image computing and computer-assisted intervention. pp. 234–241. Springer (2015)

    work page 2015

  39. [39]

    In: European Conference on Computer Vision

    Silberman, N., Hoiem, D., Kohli, P., Fergus, R.: Indoor segmentation and support inference from rgbd images. In: European Conference on Computer Vision. pp. 746–760. Springer (2012)

    work page 2012

  40. [40]

    In: Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on

    Song, S., Yu, F., Zeng, A., Chang, A.X., Savva, M., Funkhouser, T.: Semantic scene completion from a single depth image. In: Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on. pp. 190–198. IEEE (2017)

    work page 2017

  41. [41]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Tatarchenko, M., Dosovitskiy, A., Brox, T.: Octree generating networks: Efficient convolutional architectures for high-resolution 3d outputs. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 2088–2096 (2017)

    work page 2088

  42. [42]

    In: IEEE International Conference on 3D Vision (3DV) (2017)

    Uhrig, J., Schneider, N., Schneidre, L., Franke, U., Brox, T., Geiger, A.: Sparsity invariant cnns. In: IEEE International Conference on 3D Vision (3DV) (2017)

    work page 2017

  43. [43]

    In: Intelligent Robots and Systems (IROS), 2017 IEEE/RSJ International Conference on

    Varley, J., DeChant, C., Richardson, A., Ruales, J., Allen, P.: Shape completion en- abled robotic grasping. In: Intelligent Robots and Systems (IROS), 2017 IEEE/RSJ International Conference on. pp. 2442–2447. IEEE (2017)

    work page 2017

  44. [44]

    ACM Transactions on Graphics (TOG) 36(4), 72 (2017)

    Wang, P.S., Liu, Y., Guo, Y.X., Sun, C.Y., Tong, X.: O-cnn: Octree-based con- volutional neural networks for 3d shape analysis. ACM Transactions on Graphics (TOG) 36(4), 72 (2017)

    work page 2017

  45. [45]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Wu, Z., Song, S., Khosla, A., Yu, F., Zhang, L., Tang, X., Xiao, J.: 3d shapenets: A deep representation for volumetric shapes. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1912–1920 (2015)

    work page 1912

  46. [46]

    In: Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on

    Xie, S., Girshick, R., Doll´ ar, P., Tu, Z., He, K.: Aggregated residual transformations for deep neural networks. In: Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on. pp. 5987–5995. IEEE (2017)

    work page 2017

  47. [47]

    Dense 3D Object Reconstruction from a Single Depth View

    Yang, B., Rosa, S., Markham, A., Trigoni, N., Wen, H.: 3d object dense recon- struction from a single depth view. arXiv preprint arXiv:1802.00411 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  48. [48]

    ACM Transactions on Graphics (TOG) 36(4), 70 (2017)

    Yi, L., Guibas, L., Hertzmann, A., Kim, V.G., Su, H., Yumer, E.: Learning hierar- chical shape segmentation and labeling from online repositories. ACM Transactions on Graphics (TOG) 36(4), 70 (2017)

    work page 2017

  49. [49]

    Yi, L., Shao, L., Savva, M., Huang, H., Zhou, Y., Wang, Q., Graham, B., Engelcke, M., Klokov, R., Lempitsky, V., Gan, Y., Wang, P., Liu, K., Yu, F., Shui, P., Hu, B., Zhang, Y., Li, Y., Bu, R., Sun, M., Wu, W., Jeong, M., Choi, J., Kim, C., Geetchandra, A., Murthy, N., Ramu, B., Manda, B., Ramanathan, M., Kumar, G., Preetham, P., Srivastava, S., Bhugra, S...

    work page 2017

  50. [50]

    computer vision and pattern recognition (2018)

    Zhang, X., Zhou, X., Lin, M., Sun, J.: Shufflenet: An extremely efficient convolu- tional neural network for mobile devices. computer vision and pattern recognition (2018)

    work page 2018