Reproduce “NeuralFur: Animal Fur Reconstruction From Multi-View Images” 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
Vanessa Sklyarova, Berna Kabadayi, Anastasios Yiannakidis, Giorgio Becherini, Michael J. Black, Justus Thies
Researchers at the Max Planck Institute for Intelligent Systems developed NeuralFur, a system that reconstructs high-fidelity, strand-based 3D animal fur geometry from multi-view images. It integrates Vision-Language Models to provide crucial fur properties like length and growth direction, achieving superior geometric accuracy and consistency compared to previous methods, with an F-score of 46.84% on a synthetic tiger model.
Key insights
- Integrating Vision-Language Models to provide domain-specific knowledge (like fur length and growth direction) effectively compensates for the lack of large 3D animal fur datasets.
- Accurately recovering the underlying furless body shape (M_bald) before growing fur strands is critical for realistic fur placement and flow.
- Explicitly controlling fur properties like length and growth direction for specific body parts, guided by VLM insights, is paramount for generating diverse and natural-looking animal fur.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.