Research map / Point clouds / Shape and surface completion

DMF-Net: Image-Guided Point Cloud Completion with Dual-Channel Modality Fusion and Shape-Aware Upsampling Transformer

AAAI 2025 · Point clouds › Shape and surface completion · approach: Point cloud completion networks · 4 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes DMF-Net, a two-stage image-guided point cloud completion network with symmetric dual-channel modality fusion and a shape-aware upsampling transformer, outperforming XMFnet on ShapeNet-ViPC.

Problem

Single-view image-guided point cloud completion, where existing multimodal methods let the point cloud modality dominate fusion and produce non-uniform, detail-poor outputs.

Approach

Fuse image and partial point cloud features symmetrically via two cross-attention paths (each modality attends to the other) so both contribute equally to the global feature, then refine a coarse cloud with a shape-aware upsampling transformer that uses KNN geometric and feature context plus self-attention before point splitting and displacement prediction.

Key results

• Average L2-CD of 1.038 x10^-3 on ShapeNet-ViPC known categories, best across all 8 categories.
• Reduces average CD by 0.405 relative to XMFnet (1.443), a 28.1% reduction.
• Outperforms unimodal (PointAttN 2.853) and multimodal (CSDN 2.570, ViPC 3.308) baselines.

Measured effects reported in the paper

Limitations

• Completion of thin structures such as chair legs remains poor due to sparse partial input.
• Relies on ground-truth supervision that is hard to obtain in real-world settings.
• Input images are often low resolution, limiting complementary information.

Experimental setup

Datasets: ShapeNet-ViPC

Benchmarks: ShapeNet-ViPC

Metrics: L2 Chamfer distance F-Score

Baselines: AtlasNet FoldingNet PCN TopNet ECG VRC-Net PointAttN ViPC CSDN XMFnet

Builds on: XMFnet ViPC CSDN SnowflakeNet PCN SAGPool

Closest papers by meaning

• Cross-modal Learning for Image-Guided Point Cloud Shape Completion (NeurIPS 2022 · similarity 0.90)
• CDPNet: Cross-Modal Dual Phases Network for Point Cloud Completion (AAAI 2024 · similarity 0.89)
• PCDreamer: Point Cloud Completion Through Multi-view Diffusion Priors (CVPR 2025 · similarity 0.88)
• Position-Aware Guided Point Cloud Completion with CLIP Model (AAAI 2025 · similarity 0.88)
• DC-PCN: Point Cloud Completion Network with Dual-Codebook Guided Quantization (AAAI 2025 · similarity 0.87)
• SPU-IMR: Self-supervised Arbitrary-scale Point Cloud Upsampling via Iterative Mask-recovery Network (AAAI 2025 · similarity 0.87)