Search papers, labs, and topics across Lattice.
This paper introduces Kirin, a novel framework that reconstructs and generates realistic animal motion from in-the-wild video data, addressing the significant gap in high-quality motion datasets for non-human species. By creating AiM3D, the first large-scale dataset of aligned video-text-motion tuples for quadrupeds, the authors enable the development of a visual-guided motion generation model that leverages both textual and visual inputs. The resulting system not only generates accurate 3D motion sequences but also automates the rigging and animation of 3D animal models, paving the way for enhanced animation applications in various fields.
Kirin transforms raw video footage into realistic animated animal motions, bridging a critical gap in the availability of motion data for non-human species.
Understanding animal motion is fundamental to modeling animal behavior and biomechanics, yet progress in this area lags far behind human motion research due to the scarcity of high-quality motion data. While human motion can be captured in controlled environments, it is impractical for most animal species, resulting in small, domain-limited datasets that restrict downstream applications such as animation. To address this challenge, we introduce Kirin, a framework that reconstructs motion from video, learns motion priors at scale, and generates realistic motion that can be directly applied to animated assets. Using large collections of in-the-wild animal videos, we reconstruct 3D motion sequences and pair them with captions to create AiM3D, the first large-scale dataset offering aligned video-text-motion tuples for quadruped animals. Building on this dataset, we develop a visual-guided motion generation model that conditions on both text and image to guide the generation of realistic motion across diverse animal species. Finally, by leveraging an off-the-shelf image-to-3D model, we automatically rig and animate 3D meshes using generated motion, producing ready-to-render animated animals. Together, our dataset and framework establish a new foundation for large-scale, text and image conditioned animal motion generation and animation. Project page: https://kirin-ani.github.io/.
