
摘要
点云是一种重要的几何数据结构。由于其格式不规则,大多数研究人员通常将这类数据转换为规则的三维体素网格或图像集合。然而,这种处理方式会导致数据冗余,引发一系列问题。本文提出了一种新型神经网络,可直接处理点云数据,并充分尊重输入点集的排列不变性。该网络名为PointNet,为从物体分类、部件分割到场景语义解析等各类应用提供统一的架构。尽管结构简单,PointNet却具有高度的效率与有效性。实验结果表明,其性能达到甚至超越当前最先进的方法。理论上,我们进一步分析了网络所学习到的特征,以及其对输入扰动和噪声具有鲁棒性的原因。
代码仓库
GOD-GOD-Autonomous-Vehicle/self-pointnet
pytorch
GitHub 中提及
donshen/pointnet.phasedetection
pytorch
GitHub 中提及
CheesyB/cpointnet
pytorch
GitHub 中提及
zgx0534/pointnet_win
tf
GitHub 中提及
fxia22/pointnet.pytorch
pytorch
GitHub 中提及
ftdlyc/pointnet_pytorch
pytorch
GitHub 中提及
y2kmz/pointnetv2
tf
GitHub 中提及
ajertec/PointNetKeras
tf
GitHub 中提及
sarthakTUM/roofn3d
pytorch
GitHub 中提及
ajhamdi/AdvPC
tf
GitHub 中提及
charlesq34/pointnet
tf
GitHub 中提及
imanlab/deep_movement_primitives
tf
GitHub 中提及
eth212/Qmind_ModelNet10_PointNet
tf
GitHub 中提及
lingzhang1/pointnet_pytorch
pytorch
GitHub 中提及
witignite/Frustum-PointNet
tf
GitHub 中提及
alpemek/ais3d
pytorch
GitHub 中提及
minhncedutw/pointnet1_keras
tf
GitHub 中提及
bt77/pointnet
tf
GitHub 中提及
dwtstore/sfm1
tf
GitHub 中提及
coconutzs/PointNet_zs
tf
GitHub 中提及
wonderland-dsg/pointnet-grid
tf
GitHub 中提及
THHHomas/mls
pytorch
GitHub 中提及
Fragjacker/Pointcloud-grad-CAM
tf
GitHub 中提及
YiruS/pointnet_adversarial
tf
GitHub 中提及
liuch37/pointnet
pytorch
GitHub 中提及
Phimos/Paddle-PointNet
paddle
GitHub 中提及
VaanHUANG/CSCI5210HW1
tf
GitHub 中提及
LONG-9621/Extract_Point_3D
tf
GitHub 中提及
wuryantoAji/POINTNET
tf
GitHub 中提及
YanWei123/Pointnet_encoder_Foldingnet_decoder_quantization
pytorch
GitHub 中提及
brendanmallery9/entropic-barycenters
pytorch
GitHub 中提及
Dir-b/PointNet
pytorch
GitHub 中提及
Cmeijiang/PointNet_MS
mindspore
Yuto0107/pointnet
pytorch
GitHub 中提及
Taeuk-Jang/pointcompletion
pytorch
GitHub 中提及
AlfredoZermini/PointNet
GitHub 中提及
ShadowShadowWong/update-pointnet-work
pytorch
GitHub 中提及
Yang2446/pointnet
tf
GitHub 中提及
ahmed-anas/thesis-pointnet
tf
GitHub 中提及
jtpils/treegan
pytorch
GitHub 中提及
DylanWusee/pointconv_pytorch
pytorch
GitHub 中提及
smfirmin/DeepLearningFinalProject
pytorch
GitHub 中提及
amyllykoski/CycleGAN
pytorch
GitHub 中提及
vinits5/learning3d
pytorch
GitHub 中提及
timothylimyl/PointNet-Pytorch
pytorch
GitHub 中提及
opeco17/pointnet
pytorch
GitHub 中提及
Javed002/Object_detection_PointNet
pytorch
GitHub 中提及
yevgm/deepadv3d
pytorch
GitHub 中提及
KiranAkadas/My_Pointnet_v2
tf
GitHub 中提及
hz-ants/pointnet.pytorch
pytorch
GitHub 中提及
prasadsawant5/PointNet
tf
GitHub 中提及
mengxingshifen1218/pointnet.pytorch
pytorch
GitHub 中提及
ZhihaoZhu/PointNet-Implementation-Tensorflow
tf
GitHub 中提及
Moon-Xuezha/PointNet
mindspore
GitHub 中提及
SonuDileep/3-D-Object-Detection-using-PointNet
tf
GitHub 中提及
aviros/pointnet_totations
tf
GitHub 中提及
hnVfly/pointnet.mxnet
tf
GitHub 中提及
ytng001/sensemaking
tf
GitHub 中提及
soumik12345/point-cloud-segmentation
tf
GitHub 中提及
lingzhang1/pointnet_tensorflow
tf
GitHub 中提及
romaintha/pytorch_pointnet
pytorch
GitHub 中提及
xurui1217/pointnet.pytorch-master
pytorch
GitHub 中提及
saaries/PointNet
pytorch
GitHub 中提及
SBPL-Cruz/perch_pose_sampler
pytorch
GitHub 中提及
yangyucheng000/pointnet
mindspore
GitHub 中提及
KhusDM/PointNetTree
tf
GitHub 中提及
ECE685-FinalProject/3D-object-recognition
pytorch
GitHub 中提及
Lw510107/pointnet-2018.6.27-
tf
GitHub 中提及
hnmspirit/pointnet1
pytorch
GitHub 中提及
yanxp/PointNet
pytorch
GitHub 中提及
KaidongLi/tf-3d-alpha
tf
GitHub 中提及
SY-Xuan/pointnet_paddle
paddle
GitHub 中提及
freddieee/pn_6d_single
pytorch
GitHub 中提及
LONG-9621/PointNet
tf
GitHub 中提及
Mind23-2/MindCode-63
mindspore
Young98CN/pointconv_pytorch
pytorch
GitHub 中提及
Harut0726/my-pointnet-tensorflow
tf
GitHub 中提及
AndreiMoraru123/PointNet
pytorch
GitHub 中提及
nikitakaraevv/pointnet
pytorch
GitHub 中提及
Jagyan/ghost_radarpoint_detection
GitHub 中提及
BPMJG/annotated_pointnet
tf
GitHub 中提及
2023-MindSpore-1/ms-code-35
mindspore
GitHub 中提及
princeton-vl/SimpleView
pytorch
LebronGG/PointNet
tf
GitHub 中提及
merazlab/3D_Deep_Learning_Link
pytorch
GitHub 中提及
nddoshi/pointnet
pytorch
GitHub 中提及
sanantoniochili/PointCloud_KNN
tf
GitHub 中提及
AI-Guru/pointcloud_experiments
tf
GitHub 中提及
WangYueFt/dgcnn
tf
GitHub 中提及
debajyotid2/pointnet_pl
pytorch
yanx27/Pointnet
tf
GitHub 中提及
aviros/roatationPointnet
tf
GitHub 中提及
PaParaZz1/PointNet
pytorch
GitHub 中提及
ModelBunker/PointNet-TensorFlow
tf
GitHub 中提及
ali-stanford/pointnetcfd
tf
GitHub 中提及
基准测试
| 基准 | 方法 | 指标 |
|---|---|---|
| 3d-face-reconstruction-on-5 | po | 0L: po |
| 3d-part-segmentation-on-intra | PointNet | DSC (A): 49.59 DSC (V): 85.00 IoU (A): 37.75 IoU (V): 75.23 |
| 3d-part-segmentation-on-shapenet-part | PointNet | Instance Average IoU: 83.7 |
| 3d-point-cloud-classification-on-intra | PointNet | F1 score (5-fold): 0.684 |
| 3d-point-cloud-classification-on-modelnet40 | PointNet | Mean Accuracy: 86.0 Number of params: 3.47M Overall Accuracy: 89.2 |
| 3d-point-cloud-classification-on-modelnet40-c | PointNet | Error Rate: 0.283 |
| 3d-point-cloud-classification-on-scanobjectnn | PointNet | Mean Accuracy: 63.4 Overall Accuracy: 68.2 |
| 3d-semantic-segmentation-on-kitti-360 | PointNet | Model size: N/A mIoU Category: 30.42 miou: 13.07 |
| 3d-semantic-segmentation-on-semantickitti | PointNet | test mIoU: 14.6% |
| few-shot-3d-point-cloud-classification-on-1 | PointNet | Overall Accuracy: 51.97 Standard Deviation: 12.1 |
| few-shot-3d-point-cloud-classification-on-2 | PointNet | Overall Accuracy: 57.81 Standard Deviation: 15.5 |
| few-shot-3d-point-cloud-classification-on-3 | PointNet | Overall Accuracy: 46.60 Standard Deviation: 13.5 |
| few-shot-3d-point-cloud-classification-on-4 | PointNet | Overall Accuracy: 35.20 Standard Deviation: 13.5 |
| point-cloud-classification-on-pointcloud-c | PointNet | mean Corruption Error (mCE): 1.422 |
| point-cloud-segmentation-on-pointcloud-c | PointNet | mean Corruption Error (mCE): 1.178 |
| scene-segmentation-on-scannet | PointNet++ | Average Accuracy: 60.2% |
| semantic-segmentation-on-s3dis | PointNet | Number of params: N/A mAcc: 66.2 |
| semantic-segmentation-on-s3dis-area5 | PointNet | Number of params: N/A mAcc: 49.0 |
| skeleton-based-action-recognition-on-cad-120 | PointNet (5-shot) | Accuracy: 69.1% |
| supervised-only-3d-point-cloud-classification | PointNet | GFLOPs: 0.5 Number of params (M): 3.5 Overall Accuracy (PB_T50_RS): 68.0 |