hehefan / awesome-dynamic-point-cloud-analysis

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.	Method	4	8	12	16	20	24
1	MeteorNet	78.11	81.14	86.53	88.21	-	88.50
2	P4Transformer	80.13	83.17	87.54	89.56	90.24	90.94
3	PSTNet	81.14	83.50	87.88	89.90	-	91.20
4	SequentialPointNet	77.66	86.45	88.64	89.56	91.21	91.94
5	PSTNet++	81.53	83.50	88.15	90.24	-	92.68
6	Anchor-Based Spatio-Temporal Attention	80.13	87.54	89.90	91.24	-	93.03
7	PST-Transformer	81.14	83.97	88.15	91.98	-	93.73
8	Kinet	79.80	83.84	88.53	91.92	-	93.27
9	PST2 (MeteorNet + STSA)	81.14	86.53	88.55	89.22	-	-

2. NTU RBG+D 60

No.	Method	Cross Subject	Cross View
1	3DV-PointNet++	88.8	96.3
2	P4Transformer	90.2	96.4
3	PSTNet	90.5	96.5
4	PSTNet++	91.4	96.7
5	PST-Transformer	91.0	96.4
6	SequentialPointNet	90.3	97.6
7	Kinet	92.3	96.4
8	GeometryMotion-Net	92.7	98.9
9	GeometryMotion-Transformer	93.7	99.0

3. NTU RBG+D 120

No.	Method	Cross Subject	Cross Setup
1	3DV-PointNet++	82.4	93.5
2	P4Transformer	86.4	93.5
3	PSTNet	87.0	93.8
4	PSTNet++	88.6	93.8
5	PST-Transformer	87.5	94.0
6	SequentialPointNet	83.5	95.4
7	GeometryMotion-Net	90.1	93.6
8	GeometryMotion-Transformer	90.4	93.8

4. SHREC'17

No.	Method	Acc
1	PointLSTM (Min et al.)	94.7
2	Kinet	95.2

5. NvGesture

No.	Method	Acc
1	FlickerNet	86.3
2	PointLSTM (Min et al.)	87.5
3	Kinet	89.1

II. Point-level Segmentation

1. Synthia 4D

No.	Method	mIoU (3 frames)
1	MinkNet14	77.46
2	MeteorNet	81.80
3	PSTNet	82.24
4	PSTNet++	82.60
5	ASAP-Net	82.73
6	P4Transformer	83.16
7	PST-Transformer	83.95
8	Anchor-Based Spatio-Temporal Attention	84.77
9	PST2	81.86

2. SemanticKITTI

No.	Paper Title	Venue
1	SpSequenceNet: Semantic Segmentation Network on 4D Point Clouds	CVPR'20
2	LiDAR-based Recurrent 3D Semantic Segmentation with Temporal Memory Alignment	3DV'20
3	4D Panoptic LiDAR Segmentation	CVPR'21
4	LiDAR-based 4D Panoptic Segmentation via Dynamic Shifting Network	ICCV'21
5	Spatial-Temporal Transformer for 3D Point Cloud Sequences (PST2)	WACV'22

III. Other Task

No.	Paper Title	Venue
1	Fast and Furious: Real Time End-to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Net	CVPR'18
2	PointRNN: Point Recurrent Neural Network for Moving Point Cloud Processing	arXiv'19
3	Occupancy Flow: 4D Reconstruction by Learning Particle Dynamics	ICCV'19
4	Tranquil Clouds: Neural Networks for Learning Temporally Coherent Features in Point Clouds	ICLR'20
5	CaSPR: Learning Canonical Spatiotemporal Point Cloud Representations	NeurIPS'20
6	Learning Scene Dynamics from Point Cloud Sequences	IJCV'21
7	Moving Object Segmentation in 3D LiDAR Data: A Learning-based Approach Exploiting Sequential Data	RAL'21
8	PointINet: Point Cloud Frame Interpolation Network	AAAI'21
9	Self-supervised Point Cloud Prediction Using 3D Spatio-temporal Convolutional Networks	CoRL'21
10	TPU-GAN: Learning Temporal Coherence From Dynamic Point Cloud Sequences	ICLR'22
11	HOI4D: A 4D Egocentric Dataset for Category-Level Human-Object Interaction	CVPR'22
12	IDEA-Net: Dynamic 3D Point Cloud Interpolation via Deep Embedding Alignment	CVPR'22
13	Dynamic Point Cloud Compression with Cross-Sectional Approach	arXiv'22
14	Fixing Malfunctional Objects With Learned Physical Simulation and Functional Prediction	CVPR'22
15	PointMotionNet: Point-Wise Motion Learning for Large-Scale LiDAR Point Clouds Sequences	CVPR'22
16	LiDARCap: Long-range Marker-less 3D Human Motion Capture with LiDAR Point Clouds	CVPR'22
17	Point Primitive Transformer for Long-Term 4D Point Cloud Video Understanding	ECCV'22

awesome-dynamic-point-cloud-analysis's Issues

More potential related articles.

Thank you very much for your summary of point cloud + dynamic related fields, it is very beneficial for me.

Do you think Point Cloud Scene Flow, LiDAR-Based Moving Object Segmentation and 4D Panoptic LiDAR Segmentation
are subtasks under this topic? Maybe the point cloud-based scene flow tends to focus more on the motion of each point between the two frames, so I'm not quite sure if it still belongs to this topic.

Some relevant references:

• 4D Panoptic LiDAR Segmentation arXiv, github
• LMNet: Moving Object Segmentation in 3D LiDAR Data arXiv, github
• https://github.com/MaxChanger/awesome-point-cloud-scene-flow