نسخة أولية وصول مفتوح
GenIA: Generative Reconstruction with Test-Time Input Alignment
Reconstructing complete 3D object assets from monocular or sparse multi-view observations remains challenging. Generative 3D foundation models can complete object geometry beyond the observed views, but their predictions may not faithfully reproduce the observed geometry, appearance, or pose. We introduce GenIA, a fram …