Search papers, labs, and topics across Lattice.
This paper introduces Poplar, a scalable pipeline designed for the synthesis of human-centric image datasets, addressing the challenges of generating diverse and contextually appropriate images. The pipeline employs a Specify--Render--Inspect methodology, ensuring that generated images adhere to commonsense constraints and maintain a realistic photographic quality while allowing for systematic quality control. The authors successfully created Poplar-9K, a dataset of 9,401 curated image-text pairs from over 11,000 candidates, achieving a high acceptance rate of 79.9%, and provide all resources for reproducibility and customization.
Achieving a 79.9% acceptance rate, Poplar transforms the landscape of human-centric image dataset creation with its structured and quality-controlled synthesis pipeline.
Recent image generators can synthesize convincing human-centric images, yet producing a useful collection remains different from producing a single successful image. A human-centric dataset must cover varied people and contexts, avoid implausible attribute combinations, preserve an everyday photographic character, and expose quality-control decisions at scale. We present Poplar, a reproducible Specify--Render--Inspect pipeline for human-centric image dataset synthesis. Specify samples structured attributes under commonsense constraints and verbalizes them as photography-oriented prompts. Render uses a realism-adapted image generator across composition-aware aspect ratios and retries obvious technical failures. Inspect applies a single structured vision--language review to each candidate, preserving the original prompt while rejecting intrinsic image defects or material prompt mismatches. Using Poplar, we construct Poplar-9K: 9,401 curated human-centric image--text pairs retained from 11,765 reviewed candidates (79.9\% acceptance). We release the dataset together with the pipeline, configurations, immutable generation prompts, and auditable inspection records as a compact resource for building customizable human-centric collections.
