ParaScale: Scale-Calibrated Camera-Motion Transfer via a Gauge-Invariant Parallax Number

2026-06-18 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

ParaScale introduces a novel approach to camera-motion transfer, enabling creators to reuse cinematic moves across scenes with incompatible scales. It addresses the issue where naive trajectory reuse results in imperceptible or violently exaggerated motion, attributing this to the geometric fact that translation-induced image motion scales as ||T||/Z, making monocular trajectories meaningful only up to a depth-scale gauge. The core innovation is the Parallax Number Pi = ||Delta T|| / Zbar, a dimensionless, gauge-invariant descriptor proven essential for scale-faithful transfer. ParaScale functions as a plug-and-play module, reading Pi from a reference video and re-realizing it against the target scene's depth per frame, without altering rotation or requiring retraining. It integrates between pose extraction and injection. The system also introduces the Parallax Consistency Error (PCE), a scale-symmetric metric that effectively exposes scene-scale mismatch. ParaScale reduces PCE by over 3x compared to uncalibrated transfer across four orders of magnitude, maintaining visual fidelity.

Key takeaway

For computer vision engineers developing video generation tools, if you are transferring camera motion between scenes of vastly different scales, implement ParaScale. This module ensures cinematic moves are felt appropriately by preserving the gauge-invariant Parallax Number Pi, preventing imperceptible or violently exaggerated motion. It integrates seamlessly into existing pose-conditioned generators, cutting Parallax Consistency Error by over 3x without retraining.

Key insights

Scale-faithful camera-motion transfer requires preserving the gauge-invariant Parallax Number Pi, not raw trajectories.

Principles

• Translation-induced image motion scales as ||T||/Z.
• Monocular trajectories are meaningful only up to a depth-scale gauge.
• Scale-faithful transfer must preserve the Parallax Number Pi.

Method

ParaScale reads the Parallax Number Pi from a reference video and re-realizes it against the target scene's depth, per frame, leaving rotation untouched, between pose extraction and injection.

In practice

• Integrate ParaScale as a plug-and-play module.
• Use Parallax Consistency Error (PCE) to detect scene-scale mismatch.
• Drop ParaScale into any pose-conditioned generator without retraining.

Topics

• Camera-Motion Transfer
• Parallax Number
• Gauge Invariance
• Video Generation
• Computer Vision
• Pose Estimation

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Track2View: 4D-Consistent Camera-Controlled Video Generation via Paired 3D Point Tracks — Track2View uses paired 3D point tracks to enable 4D-consistent, camera-controlled video re-rendering with explicit spatiotemporal correspondences.
• Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentation — An efficient video augmentation framework converts simulated VLA data into realistic training videos, preserving task semantics and action trajectories.
• Linear Image Generation by Synthesizing Exposure Brackets — Generating linear images as exposure brackets overcomes VAE limitations for high dynamic range content.
• BodyReLux: Temporally Consistent Full-Body Video Relighting — BodyReLux enables photorealistic, temporally consistent full-body video relighting using a hybrid data capture and diffusion model with tokenized lighting control.
• OmniDirector: General Multi-Shot Camera Cloning without Cross-Paired Data — The OmniDirector framework enables general multi-shot camera cloning by encoding camera motion as grid videos and training on million-scale data.
• PhysicEdit: Teaching Image Editing Models to Respect Physics — PhysicEdit improves image editing realism by modeling physical state transitions using video data and dual-thinking mechanisms.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.