Research map / Image generation / Autoregressive and latent generation
Vision Bridge Transformer at Scale
Zhenxiong Tan, Zeqing Wang, Xingyi Yang, Songhua Liu, Xinchao 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 |
|---|---|---|---|
| use stabilized velocity matching instead of displacement matching (ablation) | ViBT · Depth-to-video | VBench Score: 0.695 -> 0.71 | Improved |
| use stabilized velocity matching instead of plain velocity matching (ablation) | ViBT · Depth-to-video | VBench Score: 0.698 -> 0.71 | Improved |
| use stabilized velocity matching instead of displacement matching (ablation) | ViBT · ImgEdit-Bench | average score: 3.50 -> 3.55 | Improved |
| use stabilized velocity matching instead of plain velocity matching (ablation) | ViBT · ImgEdit-Bench | average score: 3.36 -> 3.55 | Improved |
| lower noise scale to s=0.5 from s=1 (ablation) | ViBT · ImgEdit-Bench | average score: 3.55 -> 3.76 | Improved |
| lower noise scale to s=0.5 from s=1 (ablation) | ViBT · Depth-to-video | VBench Score: 0.709 -> 0.666 | Hurt |
| raise noise scale to s=2 from s=1 (ablation) | ViBT · Depth-to-video | VBench Score: 0.709 -> 0.711 | No clear effect |
| raise noise scale to s=2 from s=1 (ablation) | ViBT · ImgEdit-Bench | average score: 3.55 -> 3.44 | Hurt |
| raise noise scale to s=4 from s=1 (ablation) | ViBT · Depth-to-video | VBench Score: 0.709 -> 0.482 | Hurt |
| remove noise (s=0) versus s=1 (ablation) | ViBT · ImgEdit-Bench | average score: 3.55 -> 3.10 | Hurt |
| use variance-corrected sampling instead of naive Euler-Maruyama (ablation) | ViBT · image editing | visual quality | Improved |
| use shifted timestep schedule gamma=5 instead of linear (ablation) | ViBT · image editing | visual quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AnyEdit — AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025) · better on 1 of 1 shared result
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · 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
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.88)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.87)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.87)
- FrameBridge: Improving Image-to-Video Generation with Bridge Models (ICML 2025 · similarity 0.87)
- Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators (CVPR 2025 · similarity 0.87)
- Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers (ICML 2024 · 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.
