RoboCloth — Artifact Bundle (checkpoints, comparison datasets, render assets)

Companion artifact for the RoboCloth two-stage neural BRDF paper. Training, evaluation, and rendering instructions live in the code repository (milestone_instruction.md for the core pipeline, full_experiments.md for the baseline comparison experiments).

Layout

checkpoints/stage1/{Ours,Bonn,MERL}.ckpt     # pretrained stage-1 BRDF decoders, by training source
                                             #   Ours = RoboCloth, Bonn = UBOFAB19, MERL
checkpoints/stage2/RoboCloth/<mat>/          # trained stage-2 models, RoboCloth test materials
    {Ours,Bonn,MERL,PBR}_epoch<N>.ckpt       #   145 / 226 / 314 / 370 / 452
checkpoints/stage2/UBO/<mat>/                # stage-2 models, 12 held-out UBO2014 materials
checkpoints/stage2/Bonn/<mat>/               # stage-2 models, 5 held-out Bonn materials
datasets/MERL/brdfs/*.binary                 # MERL measured-BRDF database (103 materials)
datasets/UBO2014/*.btf                       # 12 held-out UBO2014 BTFs (W400xH400, L151xV151)
render_assets/onBars01_st_hp.ply             # draped-cloth mesh (qualitative renders)
render_assets/pole_spheres_v4.obj            # chrome-bar mesh (qualitative renders)

Stage-2 checkpoint filenames encode the stage-1 decoder source (Ours/Bonn/MERL) or the analytic Disney baseline (PBR), and the training epoch of the released model.

Download

pip install -U "huggingface_hub[cli]"
huggingface-cli download koalapenguin/RoboCloth --repo-type dataset \
    --include "checkpoints/*" "render_assets/*" --local-dir robocloth_artifacts