Objaverse

doi:doi:10.57702/98tc7vqj

Objaverse

The Objaverse dataset contains around 800k 3D objects. After adopting simple filter leveraging CLIP [27] to remove the objects whose rendered images are not relevant to its name, we have about 350K objects at the end. For each object, we randomly pick two elevation angles and render 32 images with the azimuth evenly distributed in [0◦, 360◦].

BibTex:

@dataset{Hang_Xu_and_Dit-Yan_Yeung_2024,
    abstract = {The Objaverse dataset contains around 800k 3D objects. After adopting simple filter leveraging CLIP [27] to remove the objects whose rendered images are not relevant to its name, we have about 350K objects at the end. For each object, we randomly pick two elevation angles and render 32 images with the azimuth evenly distributed in [0◦, 360◦].},
    author = {Hang Xu and Dit-Yan Yeung},
    doi = {10.57702/98tc7vqj},
    institution = {No Organization},
    keyword = {'3D', '3D Generation', '3D Models', '3D Object', '3D Object Generation', '3D Objects', '3D Reconstruction', '3D Scene Registration', '3D dataset', '3D generation', '3D modeling', '3D objects', '3D reconstruction', '3D shape completion', 'Annotation', 'CAD models', 'Dataset', 'Diffusion Models', 'Large Reconstruction Models', 'Meshes', 'Multi-view Images', 'NeRF', 'Objaverse', 'Object Reconstruction', 'Object Registration', 'Objects', 'Overall Text', 'Scene Understanding', 'Text-to-3D Generation', 'Textures', 'View-specific Text', 'captions', 'multi-object scenes', 'object recognition', 'physics-based positioning', 'single image', 'text-to-3D', 'text-to-3D generation'},
    month = {dec},
    publisher = {TIB},
    title = {Objaverse},
    url = {https://service.tib.eu/ldmservice/dataset/objaverse},
    year = {2024}
}