Cloth-BRDF
A Real Multi-View Multi-Light HDR Cloth Dataset

500+ cloth materials · 580 capture configurations each · 16-bit HDR PNG

Anonymous Authors1
1 Affiliations omitted for review

Paper Data (HuggingFace) Dataset Preview Citation

Overview

Cloth-BRDF teaser figure

The Capture System in Action

A real video of our capture system at work. Left: the robotic acquisition setup — two arms sweeping the camera and area light above the turntable. Right: the camera's live view as each image is captured.

Cloth-BRDF is a real-captured dataset of cloth materials designed for inverse-rendering, SVBRDF estimation, and material learning. Each material is captured under 588 (camera, light) configurations using a robot arm carrying an HDR camera that traces an arc above a rotating turntable while a second arm sweeps a calibrated area light over a plane covering the upper hemisphere. We release accurate hand-eye and turntable calibration, sample-size metadata, and a Croissant manifest at HuggingFace.

Inverse-Rendering Results

Stage-2 neural-BRDF rendering & re-lighting demo — a cloth material relit on a sphere alongside the captured scene (preview).

Dataset Preview

The dataset is split into a train set and a test set following the official lists in globals/training_list_500.txt / globals/test_list_500.txt. 20 train and 5 test materials are shown as thumbnails below; every one of the remaining materials is linked further down so you can browse all … + … without leaving this page.

Train 20 of … shown

Loading preview manifest…

Test 5 of … shown

Train all material IDs

Test all material IDs

Neural Decoder Training Pipeline

Two-stage neural BRDF decoder training pipeline

Citation

@inproceedings{anonymous2026clothbrdf,
  title  = {Cloth-BRDF: A Real Multi-View Multi-Light HDR Cloth Dataset},
  author = {Anonymous},
  booktitle = {NeurIPS Datasets and Benchmarks Track},
  year   = {2026}
}