Camera Clone, released in 2025 by the University of Hong Kong in collaboration with Zhejiang University, Kuaishou Technology, and other institutions, is a large-scale synthetic video dataset rendered using Unreal Engine 5. The related research paper is titled "...".CamCloneMaster: Enabling Reference-based Camera Control for Video GenerationThe goal is to support camera clone learning, which replicates the camera motion of a reference video while keeping the scene content unchanged, thus achieving "content reproduction + camera motion matching".

This dataset contains 391,000 videos, covering 39,100 dynamic shooting locations, each captured by 10 simultaneous cameras. Based on these viewpoint combinations, the dataset comprises 1,154,819 Triple Sets, consisting of a camera motion reference video, a content reference video, and a target video that reproduces the same camera trajectory.

Dataset feature overview:

• Generation environment: All videos were generated in 40 high-quality 3D scenes.
• Characters and Actions: Includes 66 characters and 93 actions from Fab and Mixamo.
• Camera trajectory: Multiple types of automatically generated camera trajectories are used, and the same trajectory is reused at different locations to construct pairs of trajectories.
• Camera setup: 10 synchronized cameras are deployed at each shooting location to capture multi-view data.
• Video specifications: The length is uniformly 77 frames per second, 15 FPS; the resolution is 1344×768 or 1008×576, and different generative models can be adapted through center cropping.
• File organization: Data is organized by date and track number, with each track folder containing all viewpoint videos corresponding to that track.