Research map / 3D vision / Face and shape
Dual-S3D: Hierarchical Dual-Path Selective SSM-CNN for High-Fidelity Implicit Reconstruction
Luoxi Zhang, Pragyan Shrestha, Yu Zhou, Chun Xie, Itaru Kitahara
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 |
|---|---|---|---|
| replace ViT Base with the selective SSM mixer encoder (ablation) | Dual-S3D · Efficiency on RTX 3090 | FLOPs / inference time: 1,136.5 G -> 669.8 G FLOPs; 3.51 -> 1.52 s | Improved |
| add auxiliary depth path to decouple geometry and RGB features (ablation) | Dual-S3D · 3D-FRONT | Chamfer distance: 9.48 -> 6.60 | Improved |
| use multi-stage training (geometry first, then color) instead of single stage (ablation) | Dual-S3D · 3D-FRONT | Chamfer distance: 9.91 -> 6.60 | Improved |
| add texture finetune stage with shallow RGB features (ablation) | Dual-S3D · 3D-FRONT texture | SSIM: 0.61 -> 0.83 | Improved |
| add texture finetune stage with shallow RGB features (ablation) | Dual-S3D · 3D-FRONT geometry | Chamfer distance: 6.87 -> 6.60 | No clear effect |
| use Dual-S3D instead of SSR on 3D-FRONT | single-view reconstruction · 3D-FRONT | Chamfer distance: 10.45 -> 6.60 | Improved |
| use Dual-S3D instead of SSR on Pix3D | single-view reconstruction · Pix3D | Chamfer distance: 21.79 -> 19.90 | Improved |
| train on 3D-FRONT and test on Pix3D (domain shift) | Dual-S3D · Pix3D | F-Score: 63.70 -> 49.43 | Hurt |
| Dual-S3D versus Zero-123 on texture LPIPS | single-view reconstruction · 3D-FRONT texture | LPIPS: Zero-123 0.12 vs Ours 0.14 | Hurt |
Closest papers by meaning
- Scaffold3D: SfM-Conditioned Pointmap Prediction for Multi-View 3D Reconstruction (NeurIPS 2026 · similarity 0.87)
- FlashSSM-3D: A Distilled State Space Model for Lightning-Fast Dense 3D Reconstruction (NeurIPS 2026 · similarity 0.87)
- MVD-Fusion: Single-view 3D via Depth-consistent Multi-view Generation (CVPR 2024 · similarity 0.86)
- DualPM: Dual Posed-Canonical Point Maps for 3D Shape and Pose Reconstruction (CVPR 2025 · similarity 0.86)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.85)
- Multi-view Reconstruction via SfM-guided Monocular Depth Estimation (CVPR 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.
