Text-to-3D Generation on Toys4K CL (All)
29.51CLIP SimilarityJoint Training
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Joint TrainingBackbone=TRELLIS-XL (Flow-based)2026.04 | 29.51 | 50.2 | 71.05 | |
| ReConText3DBackbone=TRELLIS-XL (Flow-based)2026.04 | 29.14 | 52.42 | 69.78 | |
| ReConText3D + L2-SPBackbone=TRELLIS-XL (Flow-based)2026.04 | 28.98 | 51.59 | 69.79 | |
| Joint TrainingBackbone=Shap-E (Diffusion-based)2026.04 | 28.49 | 83.33 | 23.63 | |
| ReConText3DBackbone=Shap-E (Diffusion-based)2026.04 | 28.44 | 83.26 | 23.17 | |
| ReConText3D + L2-SPBackbone=Shap-E (Diffusion-based)2026.04 | 28.4 | 83.84 | 23.45 | |
| Fine-tuningBackbone=Shap-E (Diffusion-based)2026.04 | 28.25 | 85.43 | 23.37 | |
| L2-SPBackbone=Shap-E (Diffusion-based)2026.04 | 27.83 | 89.92 | 23.58 | |
| Fine-tuningBackbone=TRELLIS-XL (Flow-based)2026.04 | 27.12 | 56.66 | 70.13 | |
| L2-SPBackbone=TRELLIS-XL (Flow-based)2026.04 | 27.06 | 57.65 | 68.33 |