Modelpublic

MiaTiancai/grpo-counting-model-strict-first-50

sourceHugging Facemitupdated 1y agoView on Hugging Face

0likes1downloads

Model Card

GRPO Counting Model

A Stable Diffusion 3.5-M model fine-tuned with GRPO (Generative Reinforcement Policy Optimization) method, specifically designed for generating images with precise object counting control.

Model Description

—Base Model: Stable Diffusion 3.5-M
—Training Method: GRPO (Generative Reinforcement Policy Optimization)
—Training Data: COCO80 dataset
—Key Feature: Precise control over object quantities in generated images
—Supported Range: 1-10 objects

Model Usage

1. Download the Model

python

from huggingface_hub import snapshot_download

# Download the model locally
model_path = snapshot_download(
    repo_id="MiaTiancai/grpo-counting-model-strict-first-50",
    local_dir="./grpo_counting_model"  # specify your local path
)

2. Model Inference

For inference, please refer to the Flow-GRPO repository. The repository contains all necessary code and instructions for running inference with this model.

Limitations

—Optimal performance for scenes with 1-10 objects
—May have reduced effectiveness with complex scenes
—Results depend on prompt quality and clarity

Citation

If you use this model in your research, please cite:

bibtex

@misc{grpo-counting-model,
  author = {Mia Zhao},
  title = {GRPO Counting Model: Precise Object Count Control in Image Generation},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face Model Hub},
}

License

MIT License