Idempotent Generative Network

paper: Idempotent Generative Network, https://arxiv.org/abs/2311.01462

This is a simple unofficial implementation. We trained it on the Celeba dataset.

Usage

first, download the Celeba dataset and unzip it to the data folder.

then, install the requirements:

pip install -r requirements.txt

train

modify the parameters in config.yml as your needed and run:

python train.py

generate

model checkpoint with 1000 epoch training, download here.

These are the parameters of generate.py:

• -cp: the path of checkpoint. default: ./checkpoints/model.pth
• --config: the path of config file. default: ./config.yml
• -bs: how many images to generate at once. default: 16
• --nrow: how many images are displayed in a row, only valid when steps=1. default: 4
• --steps: the times of applying model. default: 1
• --show: whether to show the generated images. default: False
• -sp: save path of the result image. default: None
• --device: the device to use. default: cuda
• --to_grayscale: whether to convert the generated images to grayscale. default: False

generate one step images:

python generate.py -cp "./checkpoints/model.pth" --config "./config.yml" -bs 128 --nrow 16 --show -sp "./result/one_step.png"

generate multi step images:

python generate.py -cp "./checkpoints/model.pth" --config "./config.yml" -bs 8 --steps 3 --show -sp "./result/three_steps.png"