Home

Awesome

Awesome Dynamic Point Cloud / Point Cloud Video / Point Cloud Sequence / 4D Point Cloud Analysis

If you find any related paper, please kindly let me know. I will keep updating the page. Thanks for your valuable contribution.

For two-frame sence flow estimation, please refer to Awesome Point Cloud Scene Flow.

I. Video/Sqeuence-level Classification

1. MSR-Action3D

No.Method4812162024
1MeteorNet78.1181.1486.5388.21-88.50
2P4Transformer80.1383.1787.5489.5690.2490.94
3PSTNet81.1483.5087.8889.90-91.20
4SequentialPointNet77.6686.4588.6489.5691.2191.94
5PSTNet++81.5383.5088.1590.24-92.68
6Anchor-Based Spatio-Temporal Attention80.1387.5489.9091.24-93.03
7PST-Transformer81.1483.9788.1591.98-93.73
8Kinet79.8083.8488.5391.92-93.27
9PST<sup>2</sup> (MeteorNet + STSA)81.1486.5388.5589.22--

2. NTU RBG+D 60

No.MethodCross SubjectCross View
13DV-PointNet++88.896.3
2P4Transformer90.296.4
3PSTNet90.596.5
4PSTNet++91.496.7
5PST-Transformer91.096.4
6SequentialPointNet90.397.6
7Kinet92.396.4
8GeometryMotion-Net92.798.9
9GeometryMotion-Transformer93.799.0

3. NTU RBG+D 120

No.MethodCross SubjectCross Setup
13DV-PointNet++82.493.5
2P4Transformer86.493.5
3PSTNet87.093.8
4PSTNet++88.693.8
5PST-Transformer87.594.0
6SequentialPointNet83.595.4
7GeometryMotion-Net90.193.6
8GeometryMotion-Transformer90.493.8

4. SHREC'17

No.MethodAcc
1PointLSTM (Min et al.)94.7
2Kinet95.2

5. NvGesture

No.MethodAcc
1FlickerNet86.3
2PointLSTM (Min et al.)87.5
3Kinet89.1

II. Point-level Segmentation

1. Synthia 4D

No.MethodmIoU (3 frames)
1MinkNet1477.46
2MeteorNet81.80
3PSTNet82.24
4PSTNet++82.60
5ASAP-Net82.73
6P4Transformer83.16
7PST-Transformer83.95
8Anchor-Based Spatio-Temporal Attention84.77
9PST<sup>2</sup>81.86

2. SemanticKITTI

No.Paper TitleVenue
1SpSequenceNet: Semantic Segmentation Network on 4D Point CloudsCVPR'20
2LiDAR-based Recurrent 3D Semantic Segmentation with Temporal Memory Alignment3DV'20
34D Panoptic LiDAR SegmentationCVPR'21
4LiDAR-based 4D Panoptic Segmentation via Dynamic Shifting NetworkICCV'21
5Spatial-Temporal Transformer for 3D Point Cloud Sequences (PST<sup>2</sup>)WACV'22

III. Other Task

No.Paper TitleVenue
1Fast and Furious: Real Time End-to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional NetCVPR'18
2PointRNN: Point Recurrent Neural Network for Moving Point Cloud ProcessingarXiv'19
3Occupancy Flow: 4D Reconstruction by Learning Particle DynamicsICCV'19
4Tranquil Clouds: Neural Networks for Learning Temporally Coherent Features in Point CloudsICLR'20
5CaSPR: Learning Canonical Spatiotemporal Point Cloud RepresentationsNeurIPS'20
6Learning Scene Dynamics from Point Cloud SequencesIJCV'21
7Moving Object Segmentation in 3D LiDAR Data: A Learning-based Approach Exploiting Sequential DataRAL'21
8PointINet: Point Cloud Frame Interpolation NetworkAAAI'21
9Self-supervised Point Cloud Prediction Using 3D Spatio-temporal Convolutional NetworksCoRL'21
10TPU-GAN: Learning Temporal Coherence From Dynamic Point Cloud SequencesICLR'22
11HOI4D: A 4D Egocentric Dataset for Category-Level Human-Object InteractionCVPR'22
12IDEA-Net: Dynamic 3D Point Cloud Interpolation via Deep Embedding AlignmentCVPR'22
13Dynamic Point Cloud Compression with Cross-Sectional ApproacharXiv'22
14Fixing Malfunctional Objects With Learned Physical Simulation and Functional PredictionCVPR'22
15PointMotionNet: Point-Wise Motion Learning for Large-Scale LiDAR Point Clouds SequencesCVPR'22
16LiDARCap: Long-range Marker-less 3D Human Motion Capture with LiDAR Point CloudsCVPR'22
17Point Primitive Transformer for Long-Term 4D Point Cloud Video UnderstandingECCV'22