Research map / Video / Video generation
Beyond Inpainting: Unleash 3D Understanding for Stable Camera-Controlled Video Re-rendering
Dongyu Chen, Yixin Guo, Shuojin Yang, Tai-Jiang Mu, Shimin Hu
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 source video concatenation to RGB-warp model (ablation) | Model (a) · Model-design ablation (human identity) | RS: 0.5954 -> 0.6463 | Improved |
| replace warped RGB with warped depth as view condition (ablation) | Model (b) · Model-design ablation (human identity) | RS: 0.6463 -> 0.6887 | Improved |
| add source video concatenation to depth view stream (ablation) | Model (c) · Model-design ablation (human identity) | RS: 0.5804 -> 0.6887 | Improved |
| add source video concatenation to depth view stream (ablation) | Model (c) · Model-design ablation (human identity) | Cs.S.: 95.17 -> 95.29 | Improved |
| add source video concatenation to depth view stream (ablation) | Model (c) · Model-design ablation (human identity) | IFS: 0.9628 -> 0.9661 | Improved |
| replace warped RGB with warped depth as view condition (ablation) | Model (b) · Model-design ablation (human identity) | Cs.B.: 94.43 -> 94.66 | Improved |
| train on multi-camera dataset (Model a vs EX-4D inpainting setup) (ablation) | Model (a) · Model-design ablation (human identity) | RS | Improved |
| train on MultiCamWarp instead of ReCamMaster dataset (ablation) | DepthDirector · Dataset ablation | R-Err: 1.88 -> 1.15 | Improved |
| train on MultiCamWarp instead of ReCamMaster dataset (ablation) | DepthDirector · Dataset ablation | T-Err: 0.12 -> 0.08 | Improved |
| train on MultiCamWarp instead of ReCamMaster dataset (ablation) | DepthDirector · Dataset ablation | Cs.Bg.: 94.97 -> 94.90 | No clear effect |
| switch base model from Wan2.2-TI2V-5B to Wan2.1-T2V-1.3B (ablation) | DepthDirector · Koala 150 videos, camera re-rendering (10 trajectories), absolute camera | RS: 0.663 -> 0.653 | Hurt |
| switch base model from Wan2.2-TI2V-5B to Wan2.1-T2V-1.3B (ablation) | DepthDirector · Koala 150 videos, camera re-rendering (10 trajectories), absolute camera | RE: 1.010 -> 1.349 | Hurt |
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.
- ReCamMaster — ReCamMaster: Camera-Controlled Generative Rendering from A Single Video (ICCV 2025) · better on 5 of 5 shared results
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
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.89)
- MultiDiff: Consistent Novel View Synthesis from a Single Image (CVPR 2024 · similarity 0.87)
- Restage4D: Reanimating Deformable 3D Reconstruction from a Single Video (NeurIPS 2025 · similarity 0.87)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.87)
- SierpinskiCam: Camera-Controlled Video Retaking with Sierpinski Triangle Pattern Cues (NeurIPS 2026 · similarity 0.87)
- Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos (NeurIPS 2025 · 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.
