Kirin: Animal Motion Generation from In-the-Wild Video
Kirin turns ordinary animal videos into text- and image-guided 3D motion for animated quadrupeds.
The paper introduces AiM3D, a large-scale dataset pairing in-the-wild animal video, captions, and reconstructed 3D motion. Its model uses both text and image conditioning to generate realistic motion across different animal species. The framework also uses an off-the-shelf image-to-3D model to rig and animate generated meshes into ready-to-render animals. HF Daily Papers' note
The paper introduces AiM3D, a large-scale dataset pairing in-the-wild animal video, captions, and reconstructed 3D motion. Its model uses both text and image conditioning to generate realistic motion across different animal species. The framework also uses an off-the-shelf image-to-3D model to rig and animate generated meshes into ready-to-render animals. HF Daily Papers' note
score 5