Research map / Video / Video generation
NormalCrafter: Learning Temporally Consistent Normals from Video Diffusion Priors
Yanrui Bin, Wenbo Hu, Haoyuan Wang, Xinya Chen, Bing Wang
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 Semantic Feature Regularization (SFR) (ablation) | NormalCrafter · Sintel (video) | mean angular error: 31.1 -> 30.7 | Improved |
| add SFR (ablation) | NormalCrafter · ScanNet (video) | mean angular error: 13.7 -> 13.3 | Improved |
| add stage 2 (pixel-space end-to-end) training (ablation) | NormalCrafter · Sintel (video) | mean angular error: 31.6 -> 30.7 | Improved |
| add stage 2 (pixel-space end-to-end) training (ablation) | NormalCrafter · ScanNet (video) | mean angular error: 14.2 -> 13.3 | Improved |
| add stage 1 (latent-space, long temporal context) training (ablation) | NormalCrafter · Sintel (video) | mean angular error: 30.7 -> 30.7 | No clear effect |
| fine-tune the VAE decoder (ablation) | NormalCrafter · Sintel (video) | mean angular error: 30.8 -> 30.7 | Improved |
| fine-tune the VAE decoder (ablation) | VAE reconstruction · normal reconstruction | mean angular error: 5.75 -> 4.07 | Improved |
| place SFR at Up1 instead of Down1 (ablation) | NormalCrafter · Sintel (video) | mean angular error: 31.0 -> 30.7 | Improved |
| place SFR at Up1 instead of Down2 (ablation) | NormalCrafter · ScanNet (video) | mean angular error: 13.6 -> 13.3 | Improved |
| NormalCrafter vs Marigold-E2E-FT on Sintel (ablation) | NormalCrafter · Sintel (video) | mean angular error: 33.5 -> 30.7 | Improved |
| NormalCrafter vs Marigold-E2E-FT on iBims | NormalCrafter · iBims | mean angular error: 15.8 -> 16.1 | No clear effect |
| NormalCrafter vs Marigold-E2E-FT on NYUv2 | NormalCrafter · NYUv2 | mean angular error: 16.2 -> 15.4 | Improved |
Later papers that compared with this method
1 later accepted paper reports NormalCrafter as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Video Generation Models are General-Purpose Vision Learners (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exploring Pre-trained Text-to-Video Diffusion Models for Referring Video Object Segmentation (ECCV 2024 · similarity 0.87)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.86)
- Alleviating Distortion in Image Generation via Multi-Resolution Diffusion Models and Time-Dependent Layer Normalization (NeurIPS 2024 · similarity 0.85)
- Learning Temporally Consistent Video Depth from Video Diffusion Priors (CVPR 2025 · similarity 0.85)
- TRIP: Temporal Residual Learning with Image Noise Prior for Image-to-Video Diffusion Models (CVPR 2024 · similarity 0.85)
- Masked Diffusion Vision-Language Models for Temporal Action Localization (NeurIPS 2026 · similarity 0.85)
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.
