JunhaoC/MADFormer-ImageNet

MADFormer-FFHQ

This repository provides checkpoints for MADFormer trained on Imagenet-256, combining autoregressive global conditioning and diffusion-based local refinement for high-resolution image synthesis.

---

📄 Paper

MADFormer: Mixed Autoregressive & Diffusion Transformers for Continuous Image Generation

---

📦 Checkpoints

• Trained for 240k steps on ImageNet-256
• Download checkpoint: ckpts.pt

---

🧪 How to Use

# TODO

💡 MADFormer supports flexible AR↔Diff trade-offs. On ImageNet-256, increasing AR layer allocation yields up to 60% FID improvements under low NFE settings.

---

📚 Citation

If you find our work useful, please cite:

@misc{chen2025madformermixedautoregressivediffusion,
      title={MADFormer: Mixed Autoregressive and Diffusion Transformers for Continuous Image Generation}, 
      author={Junhao Chen and Yulia Tsvetkov and Xiaochuang Han},
      year={2025},
      eprint={2506.07999},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2506.07999}, 
}