
Wednesday Mar 12, 2025
Arxiv paper - TrajectoryCrafter: Redirecting Camera Trajectory for Monocular Videos via Diffusion Models
In this episode, we discuss TrajectoryCrafter: Redirecting Camera Trajectory for Monocular Videos via Diffusion Models by Mark YU, Wenbo Hu, Jinbo Xing, Ying Shan. TrajectoryCrafter is a new method that precisely redirects camera paths in monocular videos by separating view changes from content generation. It uses a dual-stream conditional video diffusion model that combines point cloud renders with source videos to ensure accurate views and coherent 4D content. By training on a hybrid dataset of monocular and multi-view videos with a double-reprojection strategy, TrajectoryCrafter achieves robust performance across diverse scenes.
Comments (0)
To leave or reply to comments, please download free Podbean or
No Comments
To leave or reply to comments,
please download free Podbean App.