Reproduce “CEI-3D: Collaborative Explicit-Implicit 3D Reconstruction for Realistic and Fine-Grained Object Editing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yue Shi, Rui Shi, Yuxuan Xiong, Bingbing Ni, Wenjun Zhang
The CEI-3D framework reconstructs 3D objects from 2D images, facilitating realistic and fine-grained editing by combining explicit editable points with an implicit geometry representation. It disentangles physical attributes like lighting and material, leading to superior editing realism (lowest FID) and faster interactive modifications (16 seconds average editing time) compared to state-of-the-art methods.
Key insights
- Combining continuous implicit surface representations with discrete, editable explicit points allows for both high-fidelity rendering and precise, localized modifications without global interference.
- Disentangling physical properties like lighting and material is essential for achieving realistic edits that naturally integrate into the scene's illumination, rather than appearing as artificial overlays.
- Propagating 2D segmentation information across views to identify 3D object parts significantly improves the efficiency and intuitiveness of part-aware editing, reducing the need for extensive manual input.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.