Research map / Video / Video generation
SierpinskiCam: Camera-Controlled Video Retaking with Sierpinski Triangle Pattern Cues
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SierpinskiCam, which fills empty regions of a geometry-based camera-control proxy with a Sierpinski fractal textured dome and injects source video tokens with negative RoPE indices (NegRoPE) for camera-controlled video retaking.
Problem
Camera-controlled video retaking from a monocular source video: geometric proxies become sparse under large viewpoint changes, and source-video conditioning typically needs architectural changes or per-video adaptation.
Approach
A Sierpinski-fractal textured dome replaces black backgrounds in warped proxies to provide trackable multi-scale camera-motion cues, and source video tokens are appended to the target sequence with negative spatial RoPE indices so source and target tokens are position-disentangled without architectural changes.
Key results
- On DAVIS, SierpinskiCam reaches RotErr 0.2058 and ATE 0.6921, best among compared methods, and Dyn-MEt3R 0.7477.
- On MultiCamVideo, PSNR 19.64 vs 17.97 for ReDirector and LPIPS 0.2923 vs 0.3807.
- User study (41 participants) gives overall score 3.23 vs 2.80 for the next best method (ReDirector), on a 1-5 scale.
Surprising findings
- Plugging the Sierpinski dome into ReAngle-A-Video improves translation error and ATE but slightly worsens rotation error (0.1591 to 0.1613).
Limitations
- Degraded results with complex motion in the source video.
- Unreliable 4D reconstruction (depth, tracking) degrades output.
- Inherits limitations of the underlying video generation and 4D estimation models.
Experimental setup
Datasets: MultiCamVideo RealEstate10K DAVIS
Benchmarks: DAVIS MultiCamVideo RealEstate10K
Metrics: VBench subject consistency background consistency aesthetic quality imaging quality temporal flickering Dyn-MEt3R MEt3R RotErr TransErr ATE
Baselines: ReCamMaster TrajectoryCrafter ReDirector ReAngle-A-Video
Builds on: Wan 2.1 RoPE Flux Kontext TrajectoryCrafter ReCamMaster
Closest papers by meaning
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.86)
- IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation (NeurIPS 2025 · similarity 0.85)
- Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model (ICLR 2025 · similarity 0.85)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.84)
- Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos (NeurIPS 2025 · similarity 0.84)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.84)
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.