Research map / 3D vision / 3D generation
Benchmarking and Learning Multi-Dimensional Quality Evaluator for Text-to-3D Generation
Yujie Zhang, Bingyang Cui, Qi Yang, Zhu Li, Yiling Xu
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 adaptive quality mapping (hypernetwork-generated mapping head) (ablation) | HyperScore baseline (multi-task CLIP) · MATE-3D Overall | SRCC: 0.698 -> 0.776 | Improved |
| add conditional feature fusion (ablation) | HyperScore baseline (multi-task CLIP) · MATE-3D Overall | SRCC: 0.698 -> 0.724 | Improved |
| add conditional feature fusion on top of AQM (ablation) | HyperScore · MATE-3D Overall | SRCC: 0.776 -> 0.792 | Improved |
| add adaptive quality mapping on top of CFF (ablation) | HyperScore · MATE-3D Overall | SRCC: 0.724 -> 0.792 | Improved |
| use hypernetwork conditional model instead of separately trained networks (ablation) | HyperScore · MATE-3D Alignment | SRCC: 0.737 -> 0.739 | Improved |
| train one multi-head network without conditioning instead of separate networks (ablation) | MultiScore · MATE-3D Overall | SRCC: 0.778 -> 0.698 | Hurt |
| use element-wise multiplication instead of addition for feature aggregation (ablation) | HyperScore · MATE-3D Overall | SRCC: 0.767 -> 0.792 | Improved |
| use element-wise multiplication instead of concatenation (ablation) | HyperScore · MATE-3D Alignment | SRCC: 0.717 -> 0.739 | Improved |
| increase number of rendered viewpoints beyond 6 (ablation) | HyperScore · MATE-3D | SRCC | Mixed |
| use conditional hypernetwork evaluator instead of ImageReward | HyperScore vs best zero-shot metric · MATE-3D Alignment | SRCC: 0.651 -> 0.739 | Improved |
| fine-tune DINO v2 backbone vs zero-shot CLIPScore | DINO v2 + FT · MATE-3D Overall | SRCC: 0.510 -> 0.728 | Improved |
| use HyperScore instead of GPTEval3D | HyperScore · MATE-3D Overall | KRCC: 0.391 -> 0.603 | Improved |
Closest papers by meaning
- Gen3DEval: Using vLLMs for Automatic Evaluation of Generated 3D Objects (CVPR 2025 · similarity 0.89)
- Hi3DEval: Advancing 3D Generation Evaluation with Hierarchical Validity (NeurIPS 2025 · similarity 0.89)
- Eval3D: Interpretable and Fine-grained Evaluation for 3D Generation (CVPR 2025 · similarity 0.89)
- Retrieval-Augmented Score Distillation for Text-to-3D Generation (ICML 2024 · similarity 0.87)
- LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching (CVPR 2024 · similarity 0.87)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.86)
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.
