Research map / Point clouds / Point cloud classification
Let Images Give You More: Point Cloud Cross-Modal Training for Shape Analysis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add PointCMT training to PointNet++ | PointNet++ · ModelNet40 | OA (%): 93.4 -> 94.4 (+1.0) | Helped |
| Add PointCMT training to PointNet++ | PointNet++ · ScanObjectNN PB_T50_RS | OA (%): 79.2±0.2 -> 83.1±0.2 (+3.9) | Helped |
| Add PointCMT training to PointMLP | PointMLP · ScanObjectNN PB_T50_RS | OA (%): 85.4±0.3 -> 86.4±0.3 (+1.0) | Helped |
| Add PointCMT training to PointMLP | PointMLP · ScanObjectNN OBJ_ONLY | OA (%): 92.2±0.3 -> 93.2±0.3 (+1.0) | Helped |
| Add feature enhancement while omitting classifier enhancement (ablation) | PointNet++ · ModelNet40 | OA (%): 93.4 -> 93.8 (+0.4) | Helped |
| Add classifier enhancement while omitting feature enhancement (ablation) | PointNet++ · ModelNet40 | OA (%): 93.4 -> 94.0 (+0.6) | Helped |
| Use both feature enhancement and classifier enhancement (ablation) | PointNet++ · ScanObjectNN PB_T50_RS | OA (%): 79.4 -> 83.3 (+3.9) | Helped |
| Replace CAD-rendered images with point-cloud projections (ablation) | PointCMT · ModelNet40 | OA (%): 94.4 -> 94.0 | Hurt |
| Add color to projected images (ablation) | PointCMT · ScanObjectNN OBJ_ONLY | OA (%): 91.8 -> 90.7 | Hurt |
| Train the image feature extractor from scratch instead of pre-training it (ablation) | PointCMT · ModelNet40 | OA (%): 94.4 -> 94.2 | Hurt |
| Apply Hinton-style direct cross-modal distillation | PointNet++ · ModelNet40 | OA (%): 93.4 -> 93.1 (-0.3) | Hurt |
| Normalize image-feature mean and standard deviation to match point features before KD (ablation) | PointNet++ with Hinton KD · ModelNet40 | OA (%): 93.1 -> 93.4 | No clear effect |
Closest papers by meaning
- MM-Point: Multi-View Information-Enhanced Multi-Modal Self-Supervised 3D Point Cloud Understanding (AAAI 2024 · similarity 0.88)
- CDPNet: Cross-Modal Dual Phases Network for Point Cloud Completion (AAAI 2024 · similarity 0.88)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.87)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.87)
- DMF-Net: Image-Guided Point Cloud Completion with Dual-Channel Modality Fusion and Shape-Aware Upsampling Transformer (AAAI 2025 · similarity 0.87)
- FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.