Research map / Point clouds / Point cloud classification
ConDaFormer: Disassembled Transformer with Local Structure Enhancement for 3D Point Cloud Understanding
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 |
|---|---|---|---|
| Disassemble cubic windows into three orthogonal planes (ablation) | S3DIS point-cloud transformer · S3DIS Area 5, window 0.32m | training time: 40.8 -> 24.1 hours | Helped |
| Disassemble cubic windows into three orthogonal planes without LSE (ablation) | S3DIS point-cloud transformer · S3DIS Area 5, window 0.32m | mIoU: 71.2 -> 70.7 | Hurt |
| Share position embeddings across planes (ablation) | DaFormer · S3DIS Area 5, window 0.32m | mIoU: 69.9 -> 70.7 | Helped |
| Add LSE before attention (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 70.7 -> 72.0 | Helped |
| Add LSE after attention (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 70.7 -> 71.5 | Helped |
| Use LSE both before and after attention (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 70.7 -> 72.6 | Helped |
| Increase window size from 0.16m to 0.32m (ablation) | DaFormer · S3DIS Area 5 | mIoU: 69.7 -> 70.7 | Helped |
| Increase ConDaFormer window size from 0.32m to 0.48m or 0.64m (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 72.6 -> 72.1; 72.6 -> 71.6 | Hurt |
| Replace cRPE with Swin relative position bias (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 70.7 -> 69.6 | Hurt |
| Replace cRPE with PTv2 position encoding (ablation) | ConDaFormer · S3DIS Area 5 | mIoU: 70.7 -> 70.1 | Hurt |
| Use test-time augmentation | ConDaFormer · ScanNet v2 val | mIoU: 75.1 -> 76.0 | Helped |
| Use test-time augmentation | ConDaFormer · S3DIS Area 5 | mIoU: 72.6 -> 73.5 | Helped |
Closest papers by meaning
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.90)
- FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer (CVPR 2023 · similarity 0.90)
- LitePT: Lighter Yet Stronger Point Transformer (CVPR 2026 · similarity 0.90)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.87)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.86)
- SE(3)-bi-equivariant Transformers for Point Cloud Assembly (NeurIPS 2024 · similarity 0.86)
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.