FurE: Efficient Instance-Specific 3D Fur Reconstruction from Multi-View Images

Forum topic · 小凯 · 2026-09-30

Summary

FurE (arXiv:2609.35770) is a computer vision paper by Srinjay Sarkar, Prakhar Kaushik, Soumava Paul, and Alan Yuille on strand-based animal fur reconstruction from multi-view images. The method optimizes a root-conditioned latent field decoded into strand geometry via a PCA-based decoder, recovering a per-strand, editable animal groom. A defurred animal body is reconstructed using local fur-thickness cues from a surface-constrained Gaussian Frosting representation combined with part-based priors. To address the scarcity of animal-fur datasets, the authors show that a PCA-based decoder learned from human-hair strand data transfers to animals while significantly accelerating optimization. FurE achieves roughly 10x faster strand training than prior state-of-the-art dense per-strand optimization methods while maintaining strand fidelity, and generalizes well on both synthetic and real-world sequences.

Overview

Field: Computer Vision Authors: Srinjay Sarkar, Prakhar Kaushik, Soumava Paul, Alan Yuille arXiv: 2609.35770

Abstract (Translation)

Realistic and editable animal fur reconstruction from multi-view images is challenging due to fine-scale detail, self-occlusion and obfuscation, and — unlike human hair — the lack of animal-fur datasets. Fur usually covers most of an animal's body, with large inter-species and intra-species variability.

The authors present FurE, an efficient strand-based animal fur reconstruction method that recovers a per-strand, editable groom by optimizing a root-conditioned latent field, decoded into strand geometry via a PCA-based decoder. Key contributions:

Root-conditioned latent field: optimized per-instance and decoded into strand geometry through a PCA-based decoder, enabling per-strand editability.
Defurred body reconstruction: the animal body is reconstructed using local fur-thickness cues from a surface-constrained Gaussian Frosting representation, together with part-based priors.
Cross-domain transfer: a PCA-based decoder learned from human-hair strand data alleviates animal-data scarcity while significantly accelerating optimization.

Results

FurE is about 10x faster in strand training than the current state-of-the-art dense per-strand optimization methods while maintaining strand fidelity, and it generalizes well on both synthetic and real-world sequences.

---

*Auto-collected on 2026-09-30.*