Research map / Video / Video generation
How Far are AI-generated Videos from Simulating the 3D Visual World: A Learned 3D Evaluation Approach
Chirui Chang, Jiahui Liu, Zhengzhe Liu, Xiaoyang Lyu, Yi-Hua Huang, Xin Tao, Pengfei Wan, Di Zhang, Xiaojuan Qi
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 contrastive loss (ablation) | L3DE with concatenation fusion · Real vs synthetic detection | accuracy (%): 68.77 -> 73.14 | Improved |
| add contrastive loss (ablation) | L3DE with element-wise addition fusion · Real vs synthetic detection | accuracy (%): 66.01 -> 69.26 | Improved |
| use concatenation instead of element-wise addition (ablation) | L3DE without contrastive loss · Real vs synthetic detection | accuracy (%): 66.01 -> 68.77 | Improved |
| fuse appearance, motion and geometry instead of appearance only (ablation) | L3DE · Correlation with 3D reconstruction quality | Spearman correlation: 0.7181 -> 0.7566 | Improved |
| fuse proxies instead of appearance only (ablation) | L3DE · Correlation with human ratings | Spearman correlation: 0.5643 -> 0.6460 | Improved |
| use geometry proxy only instead of fusion (ablation) | L3DE · Correlation with 3D reconstruction quality | Spearman correlation: 0.7566 -> 0.3142 | Hurt |
| train on static plus dynamic scenes instead of static only (ablation) | L3DE · Mixed-motion test set | accuracy (%): 69.55 -> 77.70 | Improved |
| train on indoor plus outdoor instead of indoor only (ablation) | L3DE · Mixed indoor-outdoor test set | accuracy (%): 67.60 -> 76.55 | Improved |
| use L3DE instead of VBench background consistency | human rating correlation · 300 rated videos | Spearman's rho: 20.68 -> 64.84 | Improved |
| use L3DE instead of EvalCrafter | human annotation correlation · ECTV visual quality | correlation (%): 55.4 -> 67.0 | Improved |
| use L3DE instead of EvalCrafter | human annotation correlation · ECTV motion quality | correlation (%): 45.0 -> 43.6 | No clear effect |
| use L3DE instead of NPR for fake video detection | fake video detection · Fake detection average | accuracy (%): 54.25 -> 73.14 | Improved |
Closest papers by meaning
- 3DSPA: A 3D Semantic Point Autoencoder for Evaluating Video Realism (NeurIPS 2026 · similarity 0.90)
- HarmonyView: Harmonizing Consistency and Diversity in One-Image-to-3D (CVPR 2024 · similarity 0.88)
- GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025 · similarity 0.88)
- SimVS: Simulating World Inconsistencies for Robust View Synthesis (CVPR 2025 · similarity 0.87)
- 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.87)
- 4DWorldBench: A Comprehensive Evaluation Framework for 3D/4D World Generation Models (CVPR 2026 · similarity 0.87)
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.
