You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<pclass="text-lg text-gray-700 dark:text-gray-300 font-medium"><supclass="text-blue-600 font-semibold">3</sup>The Hong Kong University of Science and Technology (Guangzhou)</p>
<pclass="text-lg font-medium" style="color: #374151 !important;"><supclass="text-blue-600 font-semibold">3</sup>The Hong Kong University of Science and Technology (Guangzhou)</p>
<b>UP2You</b> reconstructs high-quality textured meshes from unconstrained photos. Our approach effectively handles extremely unconstrained photo collections by rectifying them into orthogonal multi-view images and corresponding normal maps, enabling the reconstruction of detailed 3D clothed portraits.
We present UP2You, the first tuning-free solution for reconstructing high-fidelity 3D clothed portraits from extremely unconstrained in-the-wild 2D photos.
295
293
Unlike previous approaches that require "clean" inputs (e.g., full-body images with minimal occlusions, or well-calibrated cross-view captures), UP2You directly processes raw, unstructured photographs, which may vary significantly in pose, viewpoint, cropping, and occlusion.
296
294
Instead of compressing data into tokens for slow online text-to-3D optimization, we introduce a <em>data rectifier</em> paradigm that efficiently converts unconstrained inputs into clean, orthogonal multi-view images in a single forward pass within seconds, simplifying the 3D reconstruction.
@@ -313,7 +311,7 @@ <h2>Paradigm Differences Between Previous Works and UP2You </h2>
<b>Top:</b> Previous works like PuzzleAvatar and AvatarBooth compress unconstrained photos into implicit personal tokens and DreamBooth weights through fine-tuning, then generate 3D humans via SDS optimization. <br>
318
316
<b>Bottom:</b><b>UP2You</b> directly rectifies unconstrained photo collections into orthogonal view images and normals, then reconstructs textured human meshes, achieving superior quality while reducing processing time from 4 hours to 1.5 minutes.
0 commit comments