jiayangshi/synchrotron_pixel_diffusion

Pixel Diffusion UNet – Real-world Synchrotron Dataset (DM4CT)

This repository contains the pretrained pixel-space diffusion UNet presented in the paper DM4CT: Benchmarking Diffusion Models for Computed Tomography Reconstruction.

🔗 Project Page: https://dm4ct.github.io/DM4CT/
🔗 Arxiv: https://arxiv.org/abs/2602.18589
🔗 Codebase: https://github.com/DM4CT/DM4CT

---

🔬 Model Overview

This model learns a prior over CT reconstruction images using a denoising diffusion probabilistic model (DDPM).
It operates directly in pixel space (not latent space).

• Architecture: 2D UNet (Diffusers UNet2DModel)
• Input resolution: 768 × 768
• Channels: 1 (grayscale CT slice)
• Training objective: ε-prediction (standard DDPM formulation)
• Noise schedule: Linear beta schedule
• Training dataset: Real-world Synchrotron Dataset of rocks
• Intensity normalization: Rescaled to (-1, 1)

This model is intended to be combined with data-consistency correction for CT reconstruction.

---

📊 Dataset: Real-world Synchrotron Dataset

Source: Zenodo

Preprocessing steps:

• Train/test split
• Rescale reconstructed slices to (-1, 1)
• No geometry information is embedded in the model

The model learns an unconditional image prior over CT slices.

---

🧠 Training Details

• Optimizer: AdamW
• Learning rate: 1e-4
• Hardware: NVIDIA A100 GPU
• Training script: train_pixel.py

---

🚀 Usage

You can use this model with the diffusers library as follows:

from diffusers import DDPMPipeline

# Load the pipeline
pipeline = DDPMPipeline.from_pretrained("jiayangshi/synchrotron_pixel_diffusion")

# Access the UNet model
model = pipeline.unet
model.eval()

---

Citation

@inproceedings{
shi2026dmct,
title={{DM}4{CT}: Benchmarking Diffusion Models for Computed Tomography Reconstruction},
author={Shi, Jiayang and Pelt, Dani{\"e}l M and Batenburg, K Joost},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=YE5scJekg5}
}