Stabilizing Camera-Controlled Novel View Synthesis at Inference Time

• URL: http://arxiv.org/abs/2609.03639v1
• Date: Thu, 03 Sep 2026 10:39:44 GMT
• Title: Stabilizing Camera-Controlled Novel View Synthesis at Inference Time
• Abstract summary: We show that careful control of camera motion at inference time can substantially improve the stability of camera-controlled novel view synthesis.<n>Across RealEstate10K and MegaScene, CamTrol++ improves temporal and geometric consistency, downstream 3D reconstruction quality, and generation efficiency.
• Score: 18.23602972218021
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Training-free, camera-controlled novel view synthesis from a single image using pre-trained video diffusion models often becomes unstable under large camera motion and long generation horizons. Existing approaches commonly combine several inference-time components, making it unclear which design choices are most important for stability. We show that the main source of stability is simple. Decomposing camera motion into small autoregressive steps limits per-step geometric distortion and reduces error accumulation. A controlled camera-step study shows that performance remains stable for small motions and degrades more strongly as the per-step motion approaches $18$-$20^\circ$. We further evaluate geometry-constrained spatial attention and low-frequency appearance anchoring as supporting refinements, together with an efficient registration-free warping pipeline. Across RealEstate10K and MegaScene, CamTrol++ improves temporal and geometric consistency, downstream 3D reconstruction quality, and generation efficiency over training-free baselines. The method remains effective for 56-frame generation and under substantial controlled depth corruption. These results show that careful control of camera motion at inference time can substantially improve the stability of camera-controlled novel view synthesis without retraining or modifying the diffusion backbone.

Related papers

• Directing the World: Fast Autoregressive Video Generation with Compositional Human-Camera Control [56.828569670698975]
  "Directing the World" is a fast autoregressive framework for controllable world-model video generation with compositional human-motion and camera-trajectory control.<n>Our key idea is to decouple control learning while preserving a unified autoregressive video prior.<n>For human motion control, we design a t-guided Dynamic Projection mechanism and a refined Motion-CFG strategy.
  arXiv  Detail & Related papers  (2026-06-26T11:08:09Z)
• PAS3R: Pose-Adaptive Streaming 3D Reconstruction for Long Video Sequences [6.455635287305678]
  PAS3R is a pose-adaptive streaming reconstruction framework that dynamically modulates state updates according to camera motion and scene structure.<n>We show that PAS3R significantly improves trajectory accuracy, depth estimation, and point cloud reconstruction quality in long video sequences.
  arXiv  Detail & Related papers  (2026-03-22T22:57:30Z)
• VS3R: Robust Full-frame Video Stabilization via Deep 3D Reconstruction [8.43362854806854]
  We propose VS3R, a framework that synergizes feed-forward 3D reconstruction with generative video diffusion.<n>Our pipeline jointly estimates camera parameters, depth, and masks to ensure all-scenario reliability.<n>A Dual-Stream Video Diffusion Model restores disoccluded regions and rectifies artifacts by synergizing structural guidance with semantic anchors.
  arXiv  Detail & Related papers  (2026-03-06T03:15:06Z)
• GaVS: 3D-Grounded Video Stabilization via Temporally-Consistent Local Reconstruction and Rendering [54.489285024494855]
  Video stabilization is pivotal for video processing, as it removes unwanted shakiness while preserving the original user motion intent.<n>Existing approaches, depending on the domain they operate, suffer from several issues that degrade the user experience.<n>We introduce textbfGaVS, a novel 3D-grounded approach that reformulates video stabilization as a temporally-consistent local reconstruction and rendering' paradigm.
  arXiv  Detail & Related papers  (2025-06-30T15:24:27Z)
• LightMotion: A Light and Tuning-free Method for Simulating Camera Motion in Video Generation [56.64004196498026]
  LightMotion is a light and tuning-free method for simulating camera motion in video generation.<n> operating in the latent space, it eliminates additional fine-tuning, inpainting, and depth estimation.
  arXiv  Detail & Related papers  (2025-03-09T08:28:40Z)
• Minimum Latency Deep Online Video Stabilization [77.68990069996939]
  We present a novel camera path optimization framework for the task of online video stabilization. In this work, we adopt recent off-the-shelf high-quality deep motion models for motion estimation to recover the camera trajectory. Our approach significantly outperforms state-of-the-art online methods both qualitatively and quantitatively.
  arXiv  Detail & Related papers  (2022-12-05T07:37:32Z)
• Self-Supervised Real-time Video Stabilization [100.00816752529045]
  We propose a novel method of real-time video stabilization. It transforms a shaky video to a stabilized video as if it were stabilized via gimbals in real-time.
  arXiv  Detail & Related papers  (2021-11-10T22:49:56Z)
• Neural Re-rendering for Full-frame Video Stabilization [144.9918806873405]
  We present an algorithm for full-frame video stabilization by first estimating dense warp fields. Full-frame stabilized frames can then be synthesized by fusing warped contents from neighboring frames.
  arXiv  Detail & Related papers  (2021-02-11T18:59:45Z)
• Deep Motion Blind Video Stabilization [4.544151613454639]
  This work aims to declutter this over-complicated formulation of video stabilization with the help of a novel dataset. We successfully learn motion blind full-frame video stabilization through employing strictly conventional generative techniques. Our method achieves $sim3times$ speed-up over the currently available fastest video stabilization methods.
  arXiv  Detail & Related papers  (2020-11-19T07:26:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.