RDT-3D: Resolution-Decoupled Training for High-Fidelity 3D Geometry Generation
Organizations: JD.com, Beijing, China
Abstract
High-fidelity 3D geometry generation requires maintaining global structural coherence alongside fine-grained surface details. While recent coarse-to-fine frameworks leverage Rotary Position Embeddings (RoPE) to condition local queries, we reveal a critical training-inference resolution discrepancy: coordinate guidance and geometric supervision are aligned during training, whereas inference decodes high-resolution geometry from coarse spatial coordinates. This mismatch introduces a severe distribution shift that degrades fine details. To overcome this, we propose Resolution-Decoupled Training (RDT), which anchors RoPE guidance at 512 resolution while supervising geometry at 1536, enabling robust detail extrapolation. Additionally, to eliminate high-resolution data bottlenecks, we present Dual3D-300K, a large-scale paired dataset comprising 300K watertight 3D point clouds explicitly aligned across 512 and 1536 resolutions. Extensive experiments demonstrate that RDT substantially improves sharp geometric regions, while Dual3D-300K offers a scalable benchmark for high-resolution 3D generation.
Figures & tables
| Sampling Type | Category | Res. (coarse/fine) | #Pts/sample | Normals |
|---|---|---|---|---|
| Uniform Surface | Surface Points | 300K | ✓ | |
| Curvature Surface | Surface Points | 300K | ✓ | |
| Uniform Near | Near SDF | 600K | ✓ | |
| Curvature Near | Near SDF | 300K | ✓ | |
| Free-Space | Space SDF | 400K | – | |
| Total | – | – | 1.9M | ✓ |
| Method | CD ( ) | F-Score 0.05 ( ) |
|---|---|---|
| Hunyuan3D-2.1 [ 5 ] | 0.0220 | 0.6389 |
| UltraShape [ 7 ] | 0.0185 | 0.6676 |
| TRELLIS [ 16 ] | 0.0216 | 0.6400 |
| TRELLIS.2 [ 15 ] | 0.0192 | 0.6636 |
| RDT-3D (ours) | 0.0164 | 0.6886 |
| Method | CD ( ) | F-Score ( ) | ||
|---|---|---|---|---|
| Baseline | 1536 | 1536 | 0.0184 | 0.6651 |
| Baseline with Jitter | 1536* | 1536 | 0.0187 | 0.6659 |
| Baseline with RDT | 512 | 1536 | 0.0164 | 0.6886 |
3D Shape Generation
3D shape generation creates and edits 3D objects from learned geometric priors and text, images, sketches, or structural constraints.
See all 113 papers