Original title: MeRoPE: Metric Rotary Position Embedding for Camera-Controlled Video Generation
Zhijian Qiao, Xinjiang Wang, Jiajie Chen, Haoming Huang +4 · September 2026
When computers generate videos, directing the virtual camera's movement is difficult. Existing systems track the camera with math that breaks down when the camera travels long distances. As the physical distance grows, internal numbers in the model blow up, causing errors. To solve this, researchers built MeRoPE, a new way to track camera positions. It measures the relative angles between viewing lines using rotations that keep math values from growing out of control. It turns real-world travel distances into repeating wave signals and adds a built-in guide for how points shift between views. The researchers found that MeRoPE reliably controls the camera better than earlier methods, matching both the intended turning and traveling movements across wide distances and different lenses.
Imagine you're following a group of friends in a park, and they spread out. If you only remember their absolute positions, it becomes hard to track how they relate to each other as distances grow. In this paper, video generation systems face a similar issue with camera movements, where standard methods lose accuracy. MeRoPE is like remembering your friends' positions relative to each other, keeping everything accurate no matter how far apart they move.
This helps filmmakers and developers create videos with realistic camera movements that don't glitch. It ensures generated videos match real-world camera paths more accurately, which is crucial for applications like driving simulations or virtual reality.
The abstract does not state specific limitations, but it reports results only on nuScenes and PanShot datasets. This means its performance on other types of video generation or camera movements is not established.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
MeRoPE: Metric Rotary Position Embedding for Camera-Controlled Video Generation — Zhijian Qiao, Xinjiang Wang, Jiajie Chen et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.