This repository provides the StructGen training data and final 400-case benchmark for StructGen: Disambiguating Multi-Reference Image Generation via Structured Context Modeling.
datasets/structgen_multiref/: project-owned training images and unified JSONL annotations.
datasets/mico_dere/train.jsonl: 10,665 StructGen structured-prompt annotations used with the official MICo-150K De&Re images.
benchmarks/structgen_bench/:… See the full description on the dataset page:
https://huggingface.co/datasets/jianingPeng0382/StructGen.
