Research map / Image generation / Text-to-image diffusion
AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation
Jiayin Zhu, Linlin Yang, Yicong Li, Angela Yao
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 image-conditioned source anchoring (AnchorDS with IP-Adapter) over vanilla SDS (ablation) | GaussianDreamer on SD 1.5 · T3Bench (all prompts) | quality score: 29.7 -> 30.7 | Improved |
| add filtering of unreliable source estimates (ablation) | AnchorDS (IP-Adapter) · T3Bench (all prompts) | quality score: 30.7 -> 32.8 | Improved |
| add image adapter fine-tuning (ablation) | AnchorDS (IP-Adapter) · T3Bench (all prompts) | quality score: 30.7 -> 33.3 | Improved |
| add filtering to ControlNet-conditioned AnchorDS (ablation) | AnchorDS (ControlNet) · T3Bench (all prompts) | quality score: 30.8 -> 33.2 | Improved |
| add fine-tuning to ControlNet-conditioned AnchorDS (ablation) | AnchorDS (ControlNet) · T3Bench (all prompts) | quality score: 30.8 -> 32.9 | Improved |
| add fine-tuning on top of filtering for ControlNet (ablation) | AnchorDS (ControlNet) · T3Bench (all prompts) | quality score: 33.2 -> 32.9 | No clear effect |
| add fine-tuning vs filtering for IP-Adapter on multiple-object prompts (ablation) | AnchorDS (IP-Adapter) · T3Bench (multiple objects) | quality score: 26.5 (Filter) -> 25.7 (Finetune) | Mixed |
| use AnchorDS instead of VSD | NeRF pipeline on SD 2.1 (ControlNet) · 50 complex prompts | CLIP similarity: 0.352 -> 0.369 | Improved |
| use AnchorDS (IP-Adapter) instead of VSD | NeRF pipeline on SD 1.5 · 50 complex prompts | CLIP similarity: 0.281 -> 0.334 | Improved |
| use AnchorDS instead of SDS-Bridge | NeRF pipeline on SD 1.5 · 50 complex prompts | CLIP similarity: 0.233 -> 0.334 | Improved |
| use AnchorDS instead of SDS in surrounding-object prompts (without filter or finetune) | GaussianDreamer on SD 1.5 · T3Bench (surrounding) | quality score: 26.1 -> 24.8 | Hurt |
Closest papers by meaning
- Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D Generation (ICLR 2024 · similarity 0.89)
- VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation (CVPR 2024 · similarity 0.88)
- EfficientDreamer: High-Fidelity and Robust 3D Creation via Orthogonal-view Diffusion Priors (CVPR 2024 · similarity 0.88)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models (CVPR 2024 · similarity 0.88)
- Retrieval-Augmented Score Distillation for Text-to-3D Generation (ICML 2024 · similarity 0.88)
- LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching (CVPR 2024 · similarity 0.88)
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.
