Text-to-3D Generation on Toys4K-CL Forgetting
17.36CLIP SimilarityFine-tuning
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Fine-tuningBackbone=TRELLIS-XL (Flow-based)2026.04 | 17.36 | 38.18 | 4.27 | |
| L2-SPBackbone=TRELLIS-XL (Flow-based)2026.04 | 17.23 | 40.85 | 8.98 | |
| L2-SPBackbone=Shap-E (Diffusion-based)2026.04 | 4.77 | 15.75 | 1.49 | |
| ReConText3DBackbone=TRELLIS-XL (Flow-based)2026.04 | 4.02 | 19.5 | 6.68 | |
| ReConText3D + L2-SPBackbone=TRELLIS-XL (Flow-based)2026.04 | 3.92 | 16.17 | 4.86 | |
| Fine-tuningBackbone=Shap-E (Diffusion-based)2026.04 | 3.3 | 10.43 | 0.28 | |
| ReConText3D + L2-SPBackbone=Shap-E (Diffusion-based)2026.04 | 1.46 | 4.85 | 0.55 | |
| ReConText3DBackbone=Shap-E (Diffusion-based)2026.04 | 1.43 | 3.15 | 0.67 | |
| Joint TrainingBackbone=Shap-E (Diffusion-based)2026.04 | 1.04 | 4.77 | 1.05 | |
| Joint TrainingBackbone=TRELLIS-XL (Flow-based)2026.04 | 0.1 | 7.56 | 1.18 |