we do not have a full checkpoint conversion validation, if you encounter pipeline loading failure and unsidered output, please contact me via bili_sakura@zju.edu.cn

BiliSakura/StegoGAN-ckpt

Packaged StegoGAN checkpoints from models/raw/StegoGAN.

Variants

Model	G_A/G_B compatibility
PlanIGN_layer1	compatible (auto-detected config)
PlanIGN_layer8	compatible (auto-detected config)
google_mismatch_0.25	compatible (auto-detected config)
google_mismatch_0.45	compatible (auto-detected config)
google_mismatch_0.65	compatible (auto-detected config)

Repository layout

StegoGAN-ckpt/
  {variant}/
    generator_A/
      config.json
      diffusion_pytorch_model.safetensors
    generator_B/
      config.json
      diffusion_pytorch_model.safetensors

Usage (Inference)

import numpy as np
import torch
from PIL import Image

from src.pipelines.stegogan import StegoGANPipeline

pipe = StegoGANPipeline.from_pretrained(
    "/path/to/StegoGAN-ckpt/google_mismatch_0.25",
    device="cuda",
)

img = Image.open("/path/to/input.png").convert("RGB")
x = torch.from_numpy(np.array(img)).float().permute(2, 0, 1).unsqueeze(0) / 255.0
x = x * 2 - 1
x = x.to("cuda")

out = pipe(source=x, direction="a2b", output_type="pil")
out.images[0].save("stegogan_output.png")

Notes

• generator_A/ and generator_B/ safetensors exports are included for each variant.

Citation

@inproceedings{wu2024stegogan,
  title={StegoGAN: Leveraging Steganography for Non-Bijective Image-to-Image Translation},
  author={Wu, Sidi and others},
  booktitle={CVPR},
  year={2024}
}