BoxNet: A Deep Learning Method for 2D Bounding Box Estimation from Bird's-Eye View Point Cloud

Bingbing Liu; Ehsan Nezhadarya; Yang Liu

REVIEW 4 major objections 4 minor 22 references

BoxNet: A Deep Learning Method for 2D Bounding Box Estimation from Bird's-Eye View Point Cloud

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read BoxNet shows that a simple point-order-invariant network can regress a 2D bounding box from bird's-eye-view LiDAR points, cutting car center error to 0.14 m and orientation error to 1.8 degrees on KITTI.

desk verdict A plausible but under-specified empirical study of learned 2D box geometry from object-level BEV points; the split ambiguity must be fixed before the headline numbers can be trusted. read the letter →

arxiv 1908.07085 v1 pith:DFSXGFQU submitted 2019-08-19 cs.CV

Ehsan Nezhadarya , Yang Liu , Bingbing Liu This is my paper · ORCID

classification cs.CV

keywords 2Dboundingboxestimationbird's-eyeviewpointcloudLiDARperceptiondeeplearningorientationregressiondouble-angleencodingL-shapefittingKITTIbenchmark

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

BoxNet claims that a simple deep network can replace classical L-shape fitting when estimating a 2D bounding box from bird's-eye-view LiDAR points. Given only the unordered 2D coordinates of points inside an object, the network regresses the box center, width, length, and orientation, using training data to learn typical object sizes. On the KITTI car category, this cuts average center error from about 0.39 m to 0.14 m and average orientation error from about 8.4 to 1.8 degrees, while intersection-over-union rises from 0.68 to 0.88. If the result holds, downstream modules such as tracking and data association could get faster and more accurate box geometry without hand-designed fitting rules.

What carries the argument

The load-bearing objects are the double-angle orientation encoding and the mean-relative center regression. Because heading is not estimated, the orientation lies on $(-\pi/2, \pi/2]$; the network regresses $(\cos 2\theta, \sin 2\theta)$, which is continuous at the wrap-around, and recovers $\theta = \operatorname{atan2}(\sin 2\theta, \cos 2\theta)/2$. The center head estimates $(\tilde c_x, \tilde c_y)$, the box center relative to the point-cloud mean, and the absolute center is obtained by adding the mean back; concatenating the predicted angle and size with the pooled feature vector improves the center estimate. The feature extractor uses shared MLPs and a max-pooling layer, so the prediction does not depend on point ordering, and the size head uses ReLU to output positive width and length learned from training data.

What would settle it

Reproduce the car experiments with clusters produced by an off-the-shelf detector or segmenter on KITTI instead of ground-truth boxes; if average IoU falls toward the L-shape baseline or center error exceeds roughly 0.2 m, the reported gains depend on clean ground-truth cropping.

Watch

Extended reading notes

Core claim

The central claim is that the geometry of an object — center, width, length, and orientation — can be recovered from a set of unordered 2D bird's-eye-view LiDAR points by a small regression network with three heads. The orientation head outputs $(\cos 2\theta, \sin 2\theta)$ rather than $\theta$ itself, because the bounding-box orientation is defined on $(-\pi/2, \pi/2]$ and the doubled-angle representation is continuous across the discontinuity at $\pm \pi/2$. The center head predicts the offset from the point-cloud mean, and the recovered absolute center plus learned width and length produce the box. For cars, the average center error is 0.1401 m, the orientation error is 1.8057 degrees, and the IoU is 0.8787, versus 0.3941 m, 8.4223 degrees, and 0.6825 for the best L-shape fitting baseline.

Load-bearing premise

The evaluation assumes every test input is a clean point cluster clipped from a ground-truth bounding box, with no upstream detector noise; if real clusters contain background, partial, or merged objects, the reported improvements over L-shape fitting may shrink.

Editorial extensions

If this is right

For cars, average IoU rises from 0.6825 for the best L-shape fitting variant to 0.8787 for BoxNet on the same KITTI test samples, so a trained regressor aligns boxes with true extents far better than optimization-based fitting.
Because the size head learns a width/length prior from training data, BoxNet can output the full object size even when only a partial set of points is visible, a situation L-shape fitting cannot handle because it fits only the observed points.
Point-order invariance removes the need to sort, cluster, or segment points before fitting, which simplifies use in tracking and data-association pipelines.
Shrinking the network to 1/16 of its baseline size costs less than 2% average IoU, and inference takes roughly 7.4–8.5 ms on a P100 GPU, so the accuracy gain does not require a heavy model.
The improvement is class-dependent: orientation error drops sharply for all classes, but pedestrian center error stays near 0.10 m and remains comparable to SLF, because near-square boxes give no clear primary axis.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Not tested in the paper: running BoxNet on point clusters produced by a real detector or segmenter, rather than clusters cut from ground-truth boxes; because training never sees imperfect clusters, this test would show how much of the reported gain survives upstream noise.
The double-angle encoding likely transfers to any orientation-regression problem with 180-degree symmetry, such as 2D vehicle boxes in images, since it removes angle wrap-around ambiguity without needing a classification head.
The ablation with a nearest-neighbor convolution suggests boundary points carry more regression signal than local neighbor structure; weighting or explicitly sampling boundary points could improve accuracy on sparse and heavily occluded objects.
Extending the architecture to full 3D boxes by adding height while keeping the double-angle yaw encoding is a natural next step, and the same mean-relative center trick should apply in three dimensions.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The short version: this is a modest, honestly-written paper about a narrow intermediate step — fitting a 2D bounding box to the BEV point cloud of a single object. The architecture is a simple PointNet-style regressor, and the two tricks are predicting sin(2θ), cos(2θ) instead of θ, and predicting the box center relative to the point-cloud mean. Both are sensible and the ablations support them. The empirical comparison against SLF, a non-learning L-shape fitter, is the paper's main contribution, and the reported gains for cars are large: center error 0.14 m vs 0.39 m, orientation error 1.8° vs 8.4°, IoU 0.879 vs 0.683.

The paper does several things well. The task is cleanly defined and the evaluation metrics are standard. The ablations are informative, especially the EdgeConv experiment showing that local neighbor features do not help and the network-size study showing that a much smaller network retains most of the benefit. The authors also correctly limit orientation to (−π/2, π/2] and explain why heading is not estimable from sparse BEV points. That is honest scoping.

The soft spots are real but not fatal. The most important is the train/test split. Section IV-A extracts points from every ground-truth box and then says 15,000/5,000 car samples, but never states that the split is at frame or track level. KITTI contains multiple objects per frame and the same object across consecutive frames. If the split is random over objects, near-duplicates leak across the boundary, and that would inflate the reported margins. This is a load-bearing detail for the paper's central numbers, and it should have been specified. I do not think it invalidates the method — even with some leakage, a learned regressor should beat an L-shape fitter on partially observed objects — but the magnitude of the improvement is uncertain.

Second, the input is always the ground-truth box's points. The authors acknowledge this defines the task as an intermediate module, but it means the results do not directly transfer to clusters from an upstream detector. That is a scope limitation, not a hidden flaw. Third, no error bars, no code, and the resampling to 512 points is described only briefly. For a paper whose contribution is an empirical comparison, these are missing.

The citation pattern is fine; the related work on L-shape fitting is covered and the comparison to SLF is fair. I see no circularity problem — this is straightforward supervised regression on labeled boxes.

Who is this for? Anyone working on LiDAR-based tracking or detection pipelines that need fast, accurate box refinement from object clusters. It is a useful datapoint, not a breakthrough. I would send it to review if the authors can clarify the split, add variance estimates, and release code; without those, the numbers are hard to check.

My verdict: deserves a serious referee, with the expectation of at least a major revision on the experimental reporting. If the split turns out to be frame-level, the paper is probably acceptable after a light revision.

Referee Report

4 major / 4 minor

Summary. The paper proposes BoxNet, a PointNet-style regression network that takes an unordered set of 2D bird's-eye-view (BEV) LiDAR points belonging to a single object and predicts a 2D bounding box, parameterized by center, width, length, and orientation. To handle the periodicity of an unoriented box, the orientation head regresses the double-angle vector [cos(2θ), sin(2θ)] and recovers θ by arctan2; to ease center regression, the network predicts the offset from the input point-cloud mean. The method is evaluated on KITTI 3D object data after extracting points inside ground-truth boxes, and is compared with three variants of Search-based L-shape Fitting (SLF), reporting large improvements for cars, more modest improvements for cyclists, and mixed center-error results for pedestrians. Ablations study the angle representation, the relative-center parametrization, feature concatenation, EdgeConv layers, and network width.

Significance. If the reported numbers are trustworthy, the paper gives useful evidence that a simple permutation-invariant network can outperform optimization-based L-shape fitting on ground-truth-box BEV point sets, especially when only partial object geometry is observed, and that the double-angle orientation target and mean-relative center parametrization are effective design choices. The paper is deliberately scoped as an intermediate module for downstream tracking or detection, not as a full detector, and the comparison against a published non-learning baseline on a public benchmark is appropriate. The ablation study is informative and the execution-time reporting is a useful addition. However, the central quantitative claim of "significant improvement" is not currently fully supported: the train/test split is not specified at frame or track level, no uncertainty estimates are provided, and the evaluation protocol is restricted to points extracted from ground-truth boxes.

major comments (4)

[Section IV-A, Table I] The paper must specify how the 15,000/5,000 train/test split was constructed. KITTI contains multiple objects per frame and the same object across consecutive frames, so a random object-level split can place near-identical point patterns in both training and testing. As written, the protocol is consistent with such a sample-level split, which would inflate BoxNet's reported car center error of 0.14 m, orientation error of 1.81 degrees, and IoU of 0.879, and would exaggerate the margin over SLF. The split must be performed at frame or track level, and the manuscript should state this explicitly; if the original experiments used an object-level split, they need to be rerun with a non-overlapping split.
[Section IV-D, Tables II-VII] All quantitative results are reported as single point estimates without standard deviations, confidence intervals, or the number of random seeds/restarts. Given that the central claim is that BoxNet significantly improves over SLF, the paper should report variability across at least several training runs and evaluated splits, especially for the smaller cyclist and pedestrian sets where stochastic effects are likely to be larger.
[Section IV-A] The evaluation extracts points only from the ground-truth bounding boxes of KITTI annotations. Because the paper motivates BoxNet as an intermediate step for subsequent modules such as tracking, the practical claim that it "can be applied to various scenarios" is not validated for point clusters produced by an upstream detector or segmenter, which may contain background points, partial objects, or merged objects. The authors should either test on such imperfect clusters or clearly state this as a limitation of the current study.
[Section IV-A] The resampling of each point cloud to N = 512 points is not described. The manuscript should state whether points are randomly subsampled without replacement, sampled with replacement when the cloud is smaller than 512, or resampled deterministically, and whether the same resampling is applied to the point clouds used by SLF. If SLF runs on the original point clouds while BoxNet runs on resampled ones, or vice versa, the comparison is not fair; if SLF runs on resampled clouds, the effect of resampling on the baseline's performance should be discussed.

minor comments (4)

[References] Reference [19] lists the first KITTI author as "A. Geger"; the correct spelling is "A. Geiger".
[Section IV-C, Tables IV-V] The notation for the center error is inconsistent: Equation (2) defines it as errc, but the text before Table IV refers to errd and Table V uses errd. Please unify the notation.
[Figure 3] The histograms would benefit from explicit bin widths, axis labels, and a legend in the caption or figure itself; the current text does not indicate what the vertical axis represents or how the bins were chosen.
[Section IV-F and Table VII] The word "shrinked" should be "shrunk" in the text and in Table VII.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; BoxNet is a standard supervised regression evaluated on an external benchmark.

full rationale

The paper's derivation chain is a supervised learning pipeline: BoxNet is trained on 2D BEV points extracted from KITTI ground-truth boxes to regress center offset, width, length, and double-angle orientation, then evaluated on a separate set of samples using ground-truth labels. The predicted outputs are not defined in terms of the labels, nor are any fitted parameters renamed as predictions; the loss weights from cross-validation and the ablation comparisons are ordinary hyperparameter and architecture choices. The central quantitative claim, that BoxNet outperforms SLF on center error, orientation error, and IoU, is an empirical result against held-out annotations, not a consequence of self-citation or of the network's construction. Two concerns noted by a skeptical reader — the sample-level train/test split potentially lacking frame or track separation, and the use of ground-truth-box point extraction at test time — are validity or scope limitations, not circularity, because the evaluation still compares outputs with independent ground-truth labels. No circular step is identifiable under the specified criteria.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard supervised-learning assumptions, the correctness of KITTI annotations, and the sufficiency of BEV points for box estimation. The main free parameters are loss weights and a fixed resampling size, neither of which is derived from theory. No new physical or conceptual entities are introduced.

free parameters (4)

loss weights = w0=1, w1=2, w2=1
Chosen via cross-validation (Section III-B); they balance orientation, size, and center regression terms in the total loss.
resampled point count N = 512
All object point clouds are resampled to 512 points for batch processing (Section IV-A); the resampling method is unspecified.
minimum point threshold = 30
Objects with fewer than 30 points are excluded (Section IV-A); this changes the composition of the training and testing sets.
learning rate schedule = initial 0.005, exponential decay 0.7 every 250,000 steps
Optimizer settings in Section IV-B; not swept, but they affect the final performance.

assumptions (4)

domain assumption KITTI ground-truth 3D boxes and LiDAR-to-camera transforms are accurate.
Used to extract object points and compute ground-truth 2D boxes (Section IV-A).
domain assumption Points inside the ground-truth 3D box, projected to BEV, are sufficient to infer the object's 2D bounding box.
Central modeling assumption of the task (Section III-A).
standard math The double-angle representation [cos(2θ), sin(2θ)] uniquely determines θ in the range (-π/2, π/2].
Used to recover orientation via arctan2 divided by 2 (Section III-A.1).
ad hoc to paper Resampling to 512 points preserves the bounding-box-relevant geometry.
Needed for batch processing; no analysis of its effect is provided (Section IV-A).

how reviews work

0 comments

Cite this review

Pith. "Pith review of BoxNet: A Deep Learning Method for 2D Bounding Box Estimation from Bird's-Eye View Point Cloud." pith.science (2026). https://pith.science/paper/DFSXGFQU

@misc{pith2026190807085,
  author       = {Pith},
  title        = {Pith review of: BoxNet: A Deep Learning Method for 2D Bounding Box Estimation from Bird's-Eye View Point Cloud},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DFSXGFQU}},
  note         = {Machine review of arXiv:1908.07085}
}

read the original abstract

We present a learning-based method to estimate the object bounding box from its 2D bird's-eye view (BEV) LiDAR points. Our method, entitled BoxNet, exploits a simple deep neural network that can efficiently handle unordered points. The method takes as input the 2D coordinates of all the points and the output is a vector consisting of both the box pose (position and orientation in LiDAR coordinate system) and its size (width and length). In order to deal with the angle discontinuity problem, we propose to estimate the double-angle sinusoidal values rather than the angle itself. We also predict the center relative to the point cloud mean to boost the performance of estimating the location of the box. The proposed method does not rely on the ordering of points as in many existing approaches, and can accurately predict the actual size of the bounding box based on the prior information that is obtained from the training data. BoxNet is validated using the KITTI 3D object dataset, with significant improvement compared with the state-of-the-art non-learning based methods

Figures

Figures reproduced from arXiv: 1908.07085 by the authors.

**Figure 2.** BoxNet architecture. The top sub-network extracts a network a feature vector of size 1024 from a 2D point cloud of size N. The structure of fully connected layers used to estimate the bounding box angle, size and center from the feature vector is shown in the bottom sub-network. is used as an activation function for the angle estimation head (branch). At inference time, angle is estimated using θ = arctan2(sin(2θ), … view at source ↗

**Figure 3.** Distribution of orientation error (left) and center error (right) for the car category. BoxNet has significantly improved [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Qualitative examples in bounding box estimation (best in color). (a) and (b): Car, (c) and (d): Cyclist, (e) and (f): [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: EdgeConv-based BoxNet. The first two MLP layers of the feature extraction network in baseline BoxNet were replaced with EdgeConv [22]. shrinking the network size on the average IoU. As seen, even if the network size is reduced to 1/16 of its original size, its IoU only…

**Figure 6.** Figure 6: Effect of using k nearest points in EdgeConv-based BoxNet on average IoU and execution time (in ms), for the Car category. All the experiments were conducted on a single P100 GPU with batch size of 32 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Effect of shrinking the network size on average IoU. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 20 canonical work pages

[1]

W. Luo, B. Yang, and R. Urtasun, Fast and Furious: Real Time End- to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Net, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018
[2]

B. Yang, W. Luo, and R. Urtasun, PIXOR: Real-time 3D Object Detection from Point Clouds, IEEE Conference on Computer Vision and Pattern Recognition , 2018

work page 2018
[3]

M. Ren, A. Pokrovsky, B. Yang, and R. Urtasun, SBNet: Sparse Blocks Network for Fast Inference, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018
[4]

Zhou, and O Tuzel, V oxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection, IEEE Conference on Computer Vision and Pattern Recognition, 2018

Y . Zhou, and O Tuzel, V oxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018
[5]

Charles R. Qi, W. Liu, C. Wu, H. Su, and L. J. Guibas, Frustum Point- Nets for 3D Object Detection from RGB-D Data, IEEE Conference on Computer Vision and Pattern Recognition , 2018

work page 2018
[6]

J. Ku, M. Moziﬁan, J. Lee, A. Harakeh, and S. Waslander, Joint 3D proposal generation and object detection from view aggregation, arXiv preprint arXiv: 1712.02294 , December 2017

arXiv 2017
[7]

Charles R. Qi, H. Su, K. Mo, and L. J. Guibas, PointNet: Deep Learning on Point Sets for 3D Classiﬁcation and Segmentation, IEEE Conference on Computer Vision and Pattern Recognition , 2017

work page 2017
[8]

X. Chen, H. Ma, J. Wan, B. Li, and T. Xia, Multi-View 3D Object Detection Network for Autonomous Driving, IEEE Conference on Computer Vision and Pattern Recognition , 2017

work page 2017

Show all 22 references

[9]

Himmelsbach, T

M. Himmelsbach, T. Luettel, and H.-J. Wuensche, Real-time object classiﬁcation in 3D point clouds using point feature histograms, IEEE/RSJ International Conference on Intelligent Robots and Systems , 2009

2009
[10]

H. Wang, B. Wang, B. Liu, X. Meng, and G. Yang, Pedestrian recognition and tracking using 3D LiDAR for autonomous vehicle, Robotics and Autonomous Systems , vol. 88, February 2017, pp. 71- 78

2017
[11]

Zermas, I

D. Zermas, I. Izzat, and N. Papanikolopoulos, Fast segmentation of 3D point clouds: A paradigm on LiDAR data for autonomous vehicle applications, IEEE International Conference on Robotics and Automation, 2017

2017
[12]

R. A. MacLachlan, and C. Mertz, Tracking of moving objects from a moving vehicle using a scanning laser rangeﬁnder, IEEE Intelligent Transportation Systems Conference , 2006

2006
[13]

Petrovskaya, and S

A. Petrovskaya, and S. Thrun, Model Based Vehicle Tracking for Autonomous Driving in Urban Environments, Robotics: Science and Systems, 2008

2008
[14]

X. Shen, S. Pendleton, and M. H. Ang, Efﬁcient L-shape ﬁtting of laser scanner data for vehicle pose estimation, IEEE Conference on Robotics, Automation and Mechatronics , 2015

2015
[15]

H. Zhao, Q. Zhang, M. Chiba, R. Shibasaki, J. Cui, and H. Zha, Moving Object Classiﬁcation using Horizontal Laser Scan Data, IEEE International Conference on Robotics and Automation , 2009

2009
[16]

Zhang, W

X. Zhang, W. Xu, C. Dong, and J. M. Dolan, Efﬁcient L-shape ﬁtting for vehicle detection using laser scanners, IEEE Intelligent V ehicle Symposium, 2017

2017
[17]

D. Kim, K. Jo, M. Lee, and M. Sunwoo, L-shape model switching- based precise motion tracking of moving vehicles using laser scanners, IEEE Transactions on Intelligent Transportation Systems , 19(2), 2018

2018
[18]

Mousavian, D

A. Mousavian, D. Anguelov, J. Flynn, and J Kosecka, 3D Bounding Box Estimation Using Deep Learning and Geometry, arXiv preprint arXiv: 1612.00496 , December 2016

2016 arXiv
[19]

Geger, P

A. Geger, P. Lenz, C. Stiller, and R. Urtasun, Vision meets robotics: The KITTI dataset, The International Journal of Robotics Research , 32(11), 2013

2013
[20]

X. Chen, K. Kundu, Y . Zhu, H. Ma, S. Fidler and R. Urtasun, 3D Object Proposals Using Stereo Imagery for Accurate Object Class Detection, IEEE Transactions on Pattern Analysis and Machine Intelligence, 40(5), pp. 1259-1272, May 2018

2018
[21]

X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun, Monocular 3D Object Detection for Autonomous Driving, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 2147-2156, June 2016

2016
[22]

Y . Wang, Y . Sun, Z. Liu, S. E. Sarma, M. M. Bronstein and J. M. Solomon, Dynamic Graph CNN for Learning on Point Clouds , arXiv preprint arXiv:1801.07829, 2018

2018 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

W. Luo, B. Yang, and R. Urtasun, Fast and Furious: Real Time End- to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Net, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018

[2] [2]

B. Yang, W. Luo, and R. Urtasun, PIXOR: Real-time 3D Object Detection from Point Clouds, IEEE Conference on Computer Vision and Pattern Recognition , 2018

work page 2018

[3] [3]

M. Ren, A. Pokrovsky, B. Yang, and R. Urtasun, SBNet: Sparse Blocks Network for Fast Inference, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018

[4] [4]

Zhou, and O Tuzel, V oxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection, IEEE Conference on Computer Vision and Pattern Recognition, 2018

Y . Zhou, and O Tuzel, V oxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection, IEEE Conference on Computer Vision and Pattern Recognition, 2018

work page 2018

[5] [5]

Charles R. Qi, W. Liu, C. Wu, H. Su, and L. J. Guibas, Frustum Point- Nets for 3D Object Detection from RGB-D Data, IEEE Conference on Computer Vision and Pattern Recognition , 2018

work page 2018

[6] [6]

J. Ku, M. Moziﬁan, J. Lee, A. Harakeh, and S. Waslander, Joint 3D proposal generation and object detection from view aggregation, arXiv preprint arXiv: 1712.02294 , December 2017

arXiv 2017

[7] [7]

Charles R. Qi, H. Su, K. Mo, and L. J. Guibas, PointNet: Deep Learning on Point Sets for 3D Classiﬁcation and Segmentation, IEEE Conference on Computer Vision and Pattern Recognition , 2017

work page 2017

[8] [8]

X. Chen, H. Ma, J. Wan, B. Li, and T. Xia, Multi-View 3D Object Detection Network for Autonomous Driving, IEEE Conference on Computer Vision and Pattern Recognition , 2017

work page 2017