Research map / 3D vision / 3D generation
SAM+D: Parameter-Efficient Dimensional Lifting of SAM-Family Models via Depth-Routed LoRA and Depth Shifting
YU SONG, Hao Sun, TENG SHIYU, Ikuko Nishikawa, Yen-wei Chen
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add Depth Shift Module to DRLoRA (ablation) | SAM+D · LiTS | Dice (%): 56.41 -> 59.17 | Improved |
| add depth embedding (z-embed) (ablation) | SAM+D · LiTS | Dice (%): 59.17 -> 61.05 | Improved |
| add LoRA in the mask decoder (ablation) | SAM+D · LiTS | Dice (%): 61.05 -> 63.33 | Improved |
| use 4 depth-routed experts instead of a single LoRA (ablation) | SAM+D · LiTS | Dice (%): 58.86 -> 63.33 | Improved |
| increase experts from 4 to 8 (ablation) | SAM+D · LiTS | Dice (%): 63.33 -> 63.10 | No clear effect |
| set DSM shift ratio to 0.25 instead of 0.125 (ablation) | SAM+D · LiTS | Dice (%): 61.22 -> 63.33 | Improved |
| set DSM shift ratio to 0.25 instead of 0.5 (ablation) | SAM+D · LiTS | Dice (%): 62.08 -> 63.33 | Improved |
| use LoRA rank 16 instead of 4 (ablation) | SAM+D · LiTS | Dice (%): 60.72 -> 63.33 | Improved |
| use LoRA rank 16 instead of 32 (ablation) | SAM+D · LiTS | Dice (%): 62.19 -> 63.33 | Improved |
| use DSM instead of cross-slice attention (ablation) | SAM+D · LiTS | Dice (%): 55.01 -> 63.33 | Improved |
| use DSM instead of Conv3D adapter (ablation) | SAM+D · LiTS | Dice (%): 53.62 -> 63.33 | Improved |
| use depth routing instead of content-based MoLoRA routing (ablation) | SAM+D · LiTS | Dice (%): 55.47 -> 63.33 | Improved |
Closest papers by meaning
- SAM2Object: Consolidating View Consistency via SAM2 for Zero-Shot 3D Instance Segmentation (CVPR 2025 · similarity 0.86)
- VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging (CVPR 2025 · similarity 0.86)
- UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation (NeurIPS 2026 · similarity 0.86)
- Point-SAM: Promptable 3D Segmentation Model for Point Clouds (ICLR 2025 · similarity 0.85)
- SAM-I2V: Upgrading SAM to Support Promptable Video Segmentation with Less than 0.2% Training Cost (CVPR 2025 · similarity 0.85)
- Towards a Comprehensive, Efficient and Promptable Anatomic Structure Segmentation Model Using 3D Whole-Body CT Scans (AAAI 2025 · similarity 0.85)
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.
