yzc002/LongT2IExpert

LongT2IBench: A Benchmark for Evaluating Long Text-to-Image Generation with Graph-structured Annotations

Zhichao Yang¹, Tianjiao Gu¹, Jianjie Wang¹, Feiyu Lin¹, Xiangfei Sheng¹, Pengfei Chen^1*, Leida Li^1,2*

¹School of Artificial Intelligence, Xidian University
²State Key Laboratory of Electromechanical Integrated Manufacturing of High-Performance Electronic Equipments, Xidian University

^*Corresponding author

News

• [2025-12-21] The training code has been released.
• [2025-12-09] The data and pre-trained models have been released.
• [2025-11-08] Our paper, "LongT2IBench: A Benchmark for Evaluating Long Text-to-Image Generation with Graph-structured Annotations", has been accepted for an oral presentation at AAAI 2026!

Quick Start

This guide will help you get started with the LongT2IBench inference code.

1. Installation

First, clone the repository and install the required dependencies.

git clone https://github.com/yzc-ippl/LongT2IBench.git
cd LongT2IBench
pip install -r requirements.txt

2. Download Pre-trained Weights and Dataset

Prepare Pre-trained Weights

You can download the pre-trained model weights of [LongT2IExpert] from the following link: (Baidu Netdisk)

Place the downloaded files in the weights directory.

• ./weights/LongT2IBench-checkpoints: The main model for generation and scoring.

Create the weights directory if it doesn't exist and place the files inside.

Prepare Datasets

You can download the dataset of [LongPrompt-3K] and [LongT2IBench-14K] from the following link: (Baidu Netdisk)

Place the downloaded files in the data directory.

Create the data directory if it doesn't exist and place the files inside.

LongT2IBench/
|-- weights/
|   |-- LongT2IBench-checkpoints
|   |   |-- config.json
|   |   |-- ...
|   |-- Qwen2.5-VL-7B-Instruct
|   |   |-- config.json
|   |   |-- ...
|-- data/
|   |-- imgs
|   |-- split
|   |   |-- train.json
|   |   |-- test.json
|   |   |-- val.json
|-- config.py
|-- dataset.py
|-- model.py
|-- requirements.txt
|-- README.md
|-- test_generation.py
|-- test_score.py
|-- train.py

3. Run Inference

The LongT2IExpert provides two main inference tasks: Long T2I Alignment Scoring and Long T2I Alignment Interpreting.

Long T2I Alignment Scoring

python test_score.py

Long T2I Alignment Interpreting

python test_generation.py

4. Run Training

You can run this code to train [LongT2IExpert] from start to finish.

Make sure the initially untrained weights are located at ./weights/Qwen2.5-VL-7B-Instruct :

You can download the untrained weights from the following link (Baidu Netdisk)

python train.py

Citation

If you find this work is useful, pleaes cite our paper!

@misc{yang2025longt2ibenchbenchmarkevaluatinglong,
      title={LongT2IBench: A Benchmark for Evaluating Long Text-to-Image Generation with Graph-structured Annotations}, 
      author={Zhichao Yang and Tianjiao Gu and Jianjie Wang and Feiyu Lin and Xiangfei Sheng and Pengfei Chen and Leida Li},
      year={2025},
      eprint={2512.09271},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2512.09271}, 
}