Poplar introduces a reproducible Specify–Render–Inspect pipeline for synthesizing human-centric image datasets. Specify samples structured attributes under commonsense constraints and turns them into photography-oriented prompts. Render uses a realism-adapted image generator with composition-aware aspect ratios and retries obvious technical failures. Inspect applies one structured vision-language review while retaining the original prompt, rejecting intrinsic defects and material prompt mismatches. The authors construct Poplar-9K with 9,401 curated image-text pairs from 11,765 reviewed candidates, an acceptance rate of 79.9%. The release includes the dataset, pipeline, configurations, immutable prompts, and inspection records.
No heat snapshots are available in the last 24 hours.