Jewelry Pose Dataset

Name: jewelry-dwpose-dataset
Creator: Rare Sense Inc
License: https://choosealicense.com/licenses/mit/

Dataset Structure

target: Target person images wearing jewelry
mask: Segmentation masks
dwpose: Pose maps (original unless preprocessing enabled)
prompt: Text descriptions (ring, bracelet, earring, necklace, etc.)
zoom: Macro close-up / Tight detail / Bust / Three-quarter
pose_quality: Pose quality categories (CLEAN/MILD/MODERATE)

Splits

Train: 11147 samples (ALL pose quality categories from HF_1/2/3)
Test: 25 samples (CLEAN/MILD/MODERATE pose quality only)

Pose Quality Distribution

Training (all categories): {'MILD': 1238, 'CLEAN': 6716, 'MODERATE': 594, 'SEVERE': 843, 'pocket': 210, 'EMPTY': 1546} Test (filtered): {'MILD': 10, 'CLEAN': 11, 'MODERATE': 4}

Usage

from datasets import load_dataset
ds = load_dataset("raresense/jewelry-dwpose-dataset")
train, test = ds["train"], ds["test"]

Notes

Training set includes ALL pose quality categories for maximum diversity
Test set restricted to CLEAN/MILD/MODERATE pose quality for stability
Missing files were automatically skipped during processing
Zero data leakage between train/test splits enforced
Fast processing mode enabled for efficient uploading

Datasets:

raresense
/

jewelry-dwpose-dataset