Geo-Align: Video Generation Alignment via Metric Geometry Reward
2026-05-22 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors address the challenge of editing videos by controlling the camera movements, especially when real-world data is limited. They propose a method called Geo-Align that uses reinforcement learning to improve video re-rendering by focusing on accurate camera positions and angles. Instead of needing matched video pairs for training, their approach uses a way to measure how well the generated video matches intended camera motions and adjusts accordingly. Their experiments show that Geo-Align works better than traditional supervised methods in controlling the camera and keeping the video looking good.
Camera-controlled video generationVideo-to-video re-renderingReinforcement learningScale-aware perceptual reward3D camera trajectory estimationSupervised fine-tuningSynthetic datasetsVisual fidelityOut-of-distribution generalizationMulti-view video data
Authors
Zizun Li, Haoyu Guo, Runzhe Teng, Chunhua Shen, Tong He
Abstract
Camera-controlled video generation has achieved remarkable progress in recent years. However, existing video-to-video re-rendering methods primarily rely on Supervised Fine-Tuning using synthetic datasets. At present, there is an extreme scarcity of synchronized, multi-view real-world video data. Consequently, the prevailing paradigm often exhibits limited generalization when processing out-of-distribution real-world videos, with models struggling to accurately adhere to physical scales and camera trajectories. To bridge this gap, we propose Geo-Align, the first Reinforcement Learning framework specifically designed for camera-controlled video re-rendering. Built upon a pretrained model, we optimize the model through a scale-aware perceptual reward mechanism. Specifically, we introduce a metric 3D estimator to extract precise camera trajectories from generated videos, explicitly penalizing deviations in rotation and translation. Furthermore, we meticulously designed a data pipeline strategy based on real-world conditioning videos and target camera trajectories derived from synthetic data, eliminating the reliance on paired data. Extensive experiments demonstrate that Geo-Align consistently outperforms existing supervised learning baselines in both precise camera controllability and visual fidelity, indicating the effectiveness of our method.