DAPointRWKV: Domain Adaptive Vision RWKV for Point Cloud Completion
Authors: Yinghui Li, Qianyu Zhou, Hao Yang, Richard Dazeley, Ye Zhu, Xuequan Lu
Organizations: Deakin University, Waurn Ponds, Victoria, Australia · The University of Tokyo, Tokyo, Japan · Shanghai Jiao Tong University, Shanghai, China · Deakin University, Burwood, Victoria, Australia · The University of Western Australia, Perth, Western Australia, Australia
Domain adaptive point cloud completion (DA PCC) aims to bridge the geometric discrepancy between labeled source domains and unlabeled target domains for accurate 3D shape completion. Existing methods based on CNNs, Transformers, and Mamba-like sequence models still face challenges from limited receptive fields, high computational complexity, insufficient geometric bias, and serialization sensitivity. Recently, Receptance Weighted Key Value (RWKV) provides a linear-complexity alternative for long-range sequence modeling. However, adapting RWKV to DA PCC remains challenging due to the lack of explicit geometric priors, the mismatch between recurrent modeling and coarse-to-fine generation, and unstable target-domain adaptation. To address these challenges, we propose DAPointRWKV, a novel RWKV-based framework for DA PCC. DAPointRWKV consists of three components: Geometry-Aware RWKV (GA-RWKV) Backbone, RWKV-Oriented Progressive Decoder (RWKV-PD), and Dual Structural Regularization (DSR). Specifically, GA-RWKV introduces geometry-aware recurrent representation learning, RWKV-PD bridges RWKV token representations with coarse-to-fine reconstruction through local token-guided refinement, and DSR stabilizes target-domain transfer through visible-structure anchoring and complete-shape hypothesis stabilization. Extensive experiments on synthetic and real-world benchmarks demonstrate that DAPointRWKV consistently outperforms state-of-the-art methods.
Figures & tables
Figure 1. (a) Visualization of the domain gaps among objects of the same category from different domains, where significant variations in topology exist. (b) Compared to DAPoinTr ( Li et al., 2025b ) and DAPointMamba ( Li et al., 2025c ) , DAPointRWKV achieves state-of-the-art performance on challenging benchmarks. (c) Architecture trade-off in UDA Point Cloud Completion.
Figure 2. The framework of DAPointRWKV for cross-domain point cloud completion, including three key components: (a) Geometry-Aware RWKV Backbone is designed to inject local-level geometric priors into RWKV recurrent token mixing for irregular point clouds. (b) The RWKV-Oriented Progressive Decoder is proposed to convert RWKV features into progressively refined point sets through coarse-to-fine reconstruction with local token-guided refinement. (c) Dual Structural Regularization is presented to regularize target-domain completion by enforcing observation anchoring and hypothesis stabilization.
Methods
Avg
Cabinet
Chair
Lamp
Sofa
Table
Pcl2Pcl ( Chen et al., 2019 )
92.83
57.23
43.91
157.86
63.23
141.92
ShapeInv. ( Zhang et al., 2021a )
53.21
38.54
26.30
48.57
44.02
108.60
Cycle4Comp. ( Wen et al., 2021a )
45.39
32.62
34.08
77.19
43.05
40.00
ACL-SPC ( Hong et al., 2023 )
35.97
70.12
23.87
31.75
28.74
25.38
OptDE ( Gong et al., 2022 )
28.99
28.37
21.87
29.92
37.98
26.81
DAPoinTr ( Li et al., 2025b )
22.35
18.46
17.60
27.91
23.08
24.71
Table 1. Cross-domain completion results on 3D-FUTURE. Chamfer Distance (CD) ↓ is used as the metric, and the scale factor is 104 . Lower is better.
Methods
Avg
Plane
Car
Chair
Lamp
Sofa
Table
Pcl2Pcl ( Chen et al., 2019 )
68.14
18.53
17.54
43.58
126.80
38.78
163.62
ShapeInv. ( Zhang et al., 2021a )
41.61
3.78
15.66
22.25
60.42
22.25
125.31
Cycle4Comp. ( Wen et al., 2021a )
28.65
5.77
11.85
26.67
83.34
22.82
21.47
ACL-SPC ( Hong et al., 2023 )
34.89
5.75
11.73
43.08
106.29
25.62
16.89
OptDE ( Gong et al., 2022 )
15.94
2.18
9.80
14.71
39.74
19.43
9.75
DAPoinTr ( Li et al., 2025b )
13.79
2.38
8.04
13.83
33.26
12.72
12.51
Table 2. Cross-domain completion results on ModelNet. Chamfer Distance (CD) ↓ is used as the metric, and the scale factor is 104 . Lower is better.
Methods
ScanNet
MatterPort3D
KITTI
Chair
Table
Chair
Table
Car
Pcl2Pcl ( Chen et al., 2019 )
17.3/10.1
9.1/11.8
15.9/10.5
6.0/11.8
9.2/14.1
ShapeInv. ( Zhang et al., 2021a )
3.2/10.1
3.3/11.9
3.6/10.0
3.1/11.8
2.9/13.8
Cycle4Comp. ( Wen et al., 2021a )
5.1/6.4
3.6/5.9
8.0/8.4
4.2/6.8
3.3/5.8
ACL-SPC ( Hong et al., 2023 )
1.4/4.7
1.8/5.1
1.8/4.8
2.1/4.9
2.0/4.9
OptDE ( Gong et al., 2022 )
2.6/5.5
1.9/4.6
3.0/5.5
1.9/5.3
1.6/3.5
Table 3. Cross-domain completion results on real-world scans. [UCD ↓ /UHD ↓ ] are used as the metrics, and the scale factor is 104 and 102 for UCD (Unidirectional Chamfer Distance) and UHD (Unidirectional Hausdorff Distance), respectively. A lower value of UCD or UHD is better.
Methods
GA-RWKV
RWKV-PD
OAC
HSC
Cabinet
Chair
Lamp
Sofa
Table
Avg
Baseline
21.47
16.80
23.06
22.73
19.24
20.66
DAPointRWKV
✓
19.86
16.15
22.45
21.23
19.11
19.76
✓
✓
19.31
15.88
21.31
20.58
18.43
19.10
✓
✓
✓
19.85
16.27
25.05
23.29
20.29
20.95
✓
✓
✓
19.36
15.58
14.99
19.63
14.14
16.74
✓
✓
✓
✓
19.01
15.21
13.27
19.23
13.40
16.02
Table 4. Ablation studies on each proposed module of our DAPointRWKV on 3D-FUTURE dataset. Chamfer Distance (CD) ↓ is used as the metric, and the scale factor is 104 . Lower is better.
Figure 3. Visualization comparisons between ours and state-of-the-art UDA PCC methods on 3D-FUTURE dataset.
Model
Params(M)
FLOPs(G)
Time(ms)
DAPoinTr ( Li et al., 2025b )
48.23
24.91
24.00
DAPointMamba ( Li et al., 2025c )
36.77
5.19
3.96
DAPointRWKV (Ours)
22.48
5.29
4.98
Table 5. Computational efficiency of current state-of-the-art approaches: DAPoinTr, DAPointMamba, and DAPointRWKV.
Crop Ratios
Avg
Cabinet
Chair
Lamp
Sofa
Table
45%
16.49
19.77
15.48
14.36
19.34
13.49
55%
16.02
19.01
15.21
13.27
19.23
13.40
65%
17.14
19.39
15.98
14.05
21.19
15.08
Table 6. Chamfer Distance (CD) on different crop-ratios.
The recent receptance weighted key value (RWKV) model combines RNN-style recurrence, offering a linear-complexity alternative to Transformers' quadratic self-attention for modeling global dependencies. However, when directly applied to point clouds, RWKV, originally developed for sequential text, struggles to capture local geometric structures and model spatial dependencies effectively. To address this, we propose the P-RWKV block, which bridges the gap between sequence modeling and irregular 3D geometry while preserving the efficiency advantages of RWKV. It consists of a Local Perception Expansion (LPE) component to expand contextual perception along the spatio-temporal sequence and a Spatial Context Enhancement (SCE) component to strengthen spatial awareness. To validate the effectiveness of P-RWKV for point cloud understanding, we construct PointER, a single-modality self-supervised representation learning framework whose encoder is composed of stacked P-RWKV blocks. Furthermore, we extend P-RWKV to a cross-modality setting and integrate the proposed core sub-modules into multiple architectures, demonstrating strong plug-and-play flexibility and architectural generality. Extensive experiments show that the P-RWKV block and its key sub-modules achieve competitive performance across various tasks with lower computational cost and inference latency. Code will be released upon acceptance.
Yun Liu, Xuefeng Yan, Liangliang Nan +5
School of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China · Shenzhen Institute of Research, Nanjing University of Aeronautics and Astronautics, Shenzhen, China · Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing, China +2
Existing point cloud completion methods struggle to balance high-quality reconstruction with computational efficiency. To address this, we propose PPC-MT, a novel parallel framework for point cloud completion leveraging a hybrid Mamba-Transformer architecture. Our approach introduces an innovative parallel completion strategy guided by Principal Component Analysis (PCA), which imposes a geometrically meaningful structure on unordered point clouds, transforming them into ordered sets and decomposing them into multiple subsets. These subsets are reconstructed in parallel using a multi-head reconstructor. This structured parallel synthesis paradigm significantly enhances the uniformity of point distribution and detail fidelity, while preserving computational efficiency. By integrating Mamba's linear complexity for efficient feature extraction during encoding with the Transformer's capability to model fine-grained multi-sequence relationships during decoding, PPC-MT effectively balances efficiency and reconstruction accuracy. Extensive quantitative and qualitative experiments on benchmark datasets, including PCN, ShapeNet-55/34, and KITTI, demonstrate that PPC-MT outperforms state-of-the-art methods across multiple metrics, validating the efficacy of our proposed framework.
Jie Li, Shengwei Tian, Long Yu +1
Xinjiang University, Urumqi, China · Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China
We present an image-conditioned point cloud completion approach that treats images as the primary geometric source rather than a secondary guide. To this end, we introduce an Image-to-Point (I2P) module that can reconstruct complete point clouds directly from a single RGB image, with no need for 3D inputs. Additionally, we introduce a transformer-based Point-to-Point (P2P) refinement module that uses self- and cross-attention between point tokens and image features to iteratively refine the coarse I2P output. The I2P module enables the image encoder to learn rich geometric representations, while the P2P module progressively recovers fine-grained details. Unlike existing multimodal methods that rely on auxiliary losses or fusion modules, our explicit I2P task provides a strong, geometry-aware prior based on images alone. Extensive experiments on ShapeNet-ViPC demonstrate state-of-the-art completion performance with a 12.3% relative Chamfer Distance improvement over prior methods. Code is available at: https://github.com/AzharSindhi/I2PRef.git