BootComp: Controllable Human Image Generation with Personalized Multi-Garments

This is the official implementation of the paper "Controllable Human Image Generation with Personalized Multi-Garments".

---

 

Star ⭐ us if you like it!

---

Requirements

git clone https://github.com/omnious/BootComp.git
cd BootComp

conda env create -n bootcomp python=3.10
conda activate bootcomp
pip install -r requirements.txt

Inference

Decomposition (Synthetic data generation)

You just need to specify the path to the human image from which you want to extract garments, and you're good to go.

accelerate launch data_gen.py \
--human_img_path="./example/human.jpg" \
--output_dir="decomp_output"

or, you can simply run with the script file.

sh data_gen.sh

Composition

You first need to prepare garment images and specify them in a json file following the ./example/info.json. Your json file should be as follows,

{
    "{index}": {
        "{garment category 1}": "{image path of garment1}",
        "{garment category 2}": "{image path of garment2}",
        ...
        "{garment category N}": "{image path of garmentN}",
        "text":{text prompt describing human image}
    }
}

With the prepared info_dict.json file, you can generate human images wearing multiple garments.

accelerate launch gen_composition_xl.py \
--output_dir="comp_output" \
--info_path="./example/info.json" \
--num_inference_steps=30 \
--cloth_scale=2.0 \
--guidance_scale=4.0

or, you can simply run with the script file.

sh gen_composition_xl.sh

Citation

@article{choi2024controllable,
  title={Controllable Human Image Generation with Personalized Multi-Garments},
  author={Choi, Yisol and Kwak, Sangkyung and Yu, Sihyun and Choi, Hyungwon and Shin, Jinwoo},
  journal={arXiv preprint arXiv:2411.16801},
  year={2024}
}

License

The codes and checkpoints in this repository are under the CC BY-NC-SA 4.0 license.