Research map / Video / Video generation
NÜWA: Visual Synthesis Pre-training for Neural visUal World creAtion
Chenfei Wu, Jian Liang, Lei Ji, Fan Yang, Yuejian Fang, Daxin Jiang, Nan Duan
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 T2I pre-training to T2V (ablation) | NÜWA · MSR-VTT T2V | CLIPSIM: 0.2314 -> 0.2379 | Improved |
| add T2I pre-training to T2V (ablation) | NÜWA · MSR-VTT T2V | FID-vid: 52.98 -> 53.92 | Hurt |
| add V2V pre-training to T2V (ablation) | NÜWA · MSR-VTT T2V | FID-vid: 52.98 -> 51.81 | Improved |
| add V2V pre-training to T2V (ablation) | NÜWA · MSR-VTT T2V | CLIPSIM: 0.2314 -> 0.2335 | Improved |
| add all three tasks T2V+T2I+V2V (ablation) | NÜWA · MSR-VTT T2V | FID-vid: 52.98 -> 47.68 | Improved |
| use nearby attention in decoder instead of full (ablation) | NÜWA · VSPW S2V | FID-vid: 35.21 -> 32.06 | Improved |
| use nearby attention in encoder instead of full (ablation) | NÜWA · VSPW S2V | FID-vid: 35.21 -> 33.63 | Improved |
| use nearby-sparse instead of axial-sparse attention (ablation) | NÜWA · VSPW S2V | FID-vid: 29.18 -> 27.79 | Improved |
| use nearby-sparse instead of axial-sparse attention (ablation) | NÜWA · VSPW S2V | Detected PA: 0.5957 -> 0.6085 | Improved |
| use VQ-GAN instead of VQ-VAE (ablation) | NÜWA · ImageNet reconstruction (256^2 to 16^2) | FID: 13.3 -> 6.04 | Improved |
| use more discrete tokens (16^2 to 32^2) (ablation) | NÜWA · ImageNet reconstruction | FID: 6.04 -> 2.03 | Improved |
| 336^2 to 21^2 tokens vs 256^2 to 16^2 at same F16 (ablation) | NÜWA · ImageNet reconstruction | FID: 6.04 -> 4.79 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CCVS — CCVS: Context-aware Controllable Video Synthesis (NeurIPS 2021) · 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
- NUWA-Infinity: Autoregressive over Autoregressive Generation for Infinite Visual Synthesis (NeurIPS 2022 · similarity 0.86)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.84)
- NViST: In the Wild New View Synthesis from a Single Image with Transformers (CVPR 2024 · similarity 0.84)
- IT3D: Improved Text-to-3D Generation with Explicit View Synthesis (AAAI 2024 · similarity 0.84)
- DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025 · similarity 0.84)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.83)
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.
