FurE: Efficient Instance-Specific 3D Fur Reconstruction without Animal-Fur Datasets
Srinjay Sarkar, Prakhar Kaushik, Soumava Paul, Alan Yuille
Johns Hopkins University
Published September 28, 2026 · on the feed September 30, 2026 · ▲ 8 · arXiv 2609.35770
The abstract
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. We 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. We reconstruct a defurred animal body using local fur-thickness cues from a surface-constrained Gaussian Frosting representation together with part-based priors. We further show that a PCA-based decoder learned from human-hair strand data can alleviate animal-data scarcity while enabling substantially faster optimization. FurE achieves a 10x speedup in strand training over current SOTA dense per-strand optimization while retaining strand fidelity and generalizing across synthetic and real-world sequences, with quantitative and qualitative validation despite the reduction in training time.
The Brief
1:56A short podcast summary of the paper. Written and voiced by AI.
Transcript
FurE. Efficient Instance-Specific 3D Fur Reconstruction without Animal-Fur Datasets. By Srinjay Sarkar and three others, from Johns Hopkins University. Published on the twenty-eighth of September. Its code repository is public but slight: seven source files.
Reconstructing animal fur from multi-view images, realistically and in a form that can be edited, is hard for several reasons. The detail is fine-scale, strands occlude and obscure one another, and unlike human hair there are no animal-fur datasets to learn from. Fur also covers most of an animal's body, and it varies widely both between species and within one. A method cannot lean on a learned prior for animal fur, because the data to learn it from does not exist, and it cannot treat fur as a small detail, because it is most of what the camera sees.
FurE is an efficient, strand-based reconstruction method. It recovers an editable groom, strand by strand, by optimizing a latent field conditioned on strand roots. That field is decoded into strand geometry by a decoder based on PCA. Because the output is individual strands rather than a surface, the groom can be edited afterwards.
Underneath the fur sits the body. FurE reconstructs a defurred animal body using local cues about fur thickness, taken from a Gaussian Frosting representation constrained to the surface, together with priors defined per body part. The decoder is where the data gap is handled. The authors show that a PCA-based decoder learned from human-hair strand data can ease the scarcity of animal data, and it also makes optimization substantially faster.
Strand training runs ten times faster than the current state of the art in dense per-strand optimization, while keeping strand fidelity. The method generalizes across synthetic and real-world sequences, with quantitative and qualitative validation, despite the shorter training time.
The missing dataset never gets built. Human hair stands in for it at the decoder, and the result is a faster reconstruction of fur that remains editable down to the single strand.
Paper Quest · 7 chapters · about 39 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Fur covers nearly everything, and nobody has a dataset of it. That is the wall.
You are rebuilding a furry animal from multi-view photos, and artists want editable strands, not a blob. Strands hide one another, fur varies across species and body parts, and it covers almost everything the camera sees.
The road ahead
- 1 · The problem ~6 min Rebuilding animal fur as editable strands from photos is hard: strands hide each other, fur varies everywhere, there is no fur data, and per-strand fitting is slow.
- 2 · What came before ~3 min
- 3 · The key idea ~8 min Find the skin beneath the fur, give each root a small code, and let a strand decoder learned from human hair turn codes into strands.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min Learn a strand decoder from human hair, estimate the skin under the fur, then optimize per-root codes that decode into strands.
- 6 · The experiment ~6 min You’ll guess: How much faster is FurE's strand training than current dense per-strand optimization?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.