Research map / Point clouds / Point cloud classification
Instance-aware Dynamic Prompt Tuning for Pre-trained Point Cloud Models
Yaohua Zha, Jinpeng Wang, Tao Dai, Bin Chen, Zhi Wang, Shu-Tao Xia
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace static VPT prompts with instance-aware dynamic prompt generation (ablation) | Point-MAE prompt tuning · ScanObjectNN OBJ_BG | accuracy: 90.19 -> 92.48 | Improved |
| increase number of static VPT prompts from 10 to 156 (ablation) | VPT-Deep · ScanObjectNN OBJ_BG | accuracy: 89.98 -> 90.19 | No clear effect |
| use EdgeConv graph prompt module instead of MLP or Transformer (ablation) | IDPT · ScanObjectNN OBJ_BG | accuracy: 91.43 -> 92.48 | Improved |
| add IDPT instead of full fine-tuning | Point-MAE-dagger · ScanObjectNN PB_T50_RS | accuracy: 88.41 -> 88.51 | Improved |
| IDPT instead of full fine-tuning on ACT | ACT-dagger · ScanObjectNN PB_T50_RS | accuracy: 88.21 -> 87.65 | Hurt |
| IDPT instead of full fine-tuning on Point-MAE | Point-MAE · ScanObjectNN PB_T50_RS | accuracy: 85.18 -> 84.94 | Hurt |
| IDPT instead of full fine-tuning | Point-MAE · ModelNet40 | accuracy: 93.8 -> 94.4 | Improved |
| IDPT instead of static VPT on part segmentation | ACT · ShapeNetPart | mIoU_I: 85.4 -> 85.9 | Improved |
| prompt tuning versus full fine-tuning on part segmentation | ACT · ShapeNetPart | mIoU_I: 86.1 -> 85.9 | Hurt |
| insert dynamic prompt into deeper Transformer layers (ablation) | IDPT · ScanObjectNN OBJ_BG and OBJ_ONLY | accuracy | Improved |
| insert shared-parameter dynamic prompt in all layers (ablation) | IDPT · ScanObjectNN | accuracy | Hurt |
| add IDPT prompt versus head tuning only (ablation) | Point-MAE · ScanObjectNN OBJ_BG | accuracy: 87.40 -> 91.43 | Improved |
Closest papers by meaning
- GAPrompt: Geometry-Aware Point Cloud Prompt for 3D Vision Model (ICML 2025 · similarity 0.91)
- Dynamic Adapter Meets Prompt Tuning: Parameter-Efficient Transfer Learning for Point Cloud Analysis (CVPR 2024 · similarity 0.90)
- Exploring Vision Semantic Prompt for Efficient Point Cloud Understanding (ICML 2025 · similarity 0.90)
- Point-PEFT: Parameter-Efficient Fine-Tuning for 3D Pre-trained Models (AAAI 2024 · similarity 0.89)
- Point-PRC: A Prompt Learning Based Regulation Framework for Generalizable Point Cloud Analysis (NeurIPS 2024 · similarity 0.89)
- Point Cloud Pre-training with Diffusion Models (CVPR 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.
