The Code Repository for "CCDM: Continuous Conditional Diffusion Models for Image Generation"

[UPDATE! 2025-07-13] We corrected a coding error that inadvertently allowed CCDM and CcDPM to utilize additional training samples on RC-49, though fortunately our primary conclusion about CCDM's substantial superiority remains unchanged.

[UPDATE! 2025-02-26] We offer a unified code repository located at ./CCDM/CCDM_unified, which supports training CcDPM, CCDM, and DMD2-M on RC-49, Cell-200, UTKFace, and Steering Angle. The original code repository, containing the initial version of CCDM, is now archived in ./CCDM/CCDM_vanilla. Detailed training and sampling setups are documented in ./setup_details.pdf.

This repository provides the source codes for the experiments in our papers for CCDMs.
If you use this code, please cite

@misc{ding2024ccdm,
      title={{CCDM}: Continuous Conditional Diffusion Models for Image Generation}, 
      author={Xin Ding and Yongwei Wang and Kao Zhang and Z. Jane Wang},
      year={2024},
      eprint={2405.03546},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Some Illustrative Figures

Illustration of the CCGM task with sample images from the UTKFace and Steering Angle datasets.

The overall workflow of CCDMs

Software Requirements

Here, we provide a list of crucial software environments and python packages employed in the conducted experiments. Please note that we use different computational platforms for our experiments.

For computing NIQE scores (Support both Windows and Linux):

Item	Version	Item	Version
OS	Win11	OS	Linux
Python	3.11.3	Python	3.10.12
Matlab	2023a	Matlab	2021b

For implementing CCDM (Support both Windows and Linux):

Item	Version	Item	Version
OS	Win11	OS	Linux
Python	3.11.3	Python	3.10.12
CUDA	11.8	CUDA	12.1
numpy	1.23.5	numpy	1.26.4
torch	2.0.1	torch	2.2.1
torchvision	0.15.2	torchvision	0.17.1
Pillow	9.5.0	Pillow	9.0.1
accelearate	0.20.3	accelearate	0.27.2

For implementing ReACGAN, ADCGAN, ADM-G, and CFG (Support Windows):

Item	Version
OS	Win11
Python	3.11.3
CUDA	11.8
numpy	1.23.5
torch	2.0.1
torchvision	0.15.2
Pillow	9.5.0
accelearate	0.20.3
wandb	0.15.7

For implementing CcGAN and Dual-NDA (Support Linux):

Item	Version
OS	Linux
Python	3.9
CUDA	11.4
numpy	1.23.0
torch	1.12.1
torchvision	0.13.1
Pillow	8.4.0
accelearate	0.18.0

Datasets

We use the preprocessed datasets provided by Ding et. al. (2023).

The RC-49 Dataset (h5 file)

Download the following h5 file and put it in ./datasets/RC-49.

RC-49 (64x64)

RC-49_64x64_OneDrive_link
RC-49_64x64_BaiduYun_link

The preprocessed UTKFace Dataset (h5 file)

Download the following h5 files and put them in ./datasets/UTKFace.

The Steering Angle dataset (h5 file)

Download the following h5 files and put them in ./datasets/SteeringAngle.

Steering Angle (64x64)

SteeringAngle_64x64_OneDrive_link
SteeringAngle_64x64_BaiduYun_link

Steering Angle (128x128)

SteeringAngle_128x128_OneDrive_link
SteeringAngle_128x128_BaiduYun_link

The Cell-200 Dataset (h5 file)

Download the following h5 file and put it in ./datasets/Cell-200.

Cell-200 (64x64)

Cell-200_64x64_OneDrive_link
Cell-200_64x64_BaiduYun_link

Preparation (Required!)

Please download the zip file from either OneDrive or BaiduYun and extract its contents into the ./CCDM directory. The zip archive contains the required checkpoints for the ILI's embedding networks and covariance embedding networks, as well as the corresponding checkpoints for the evaluation models associated with each individual experiment.

Training

Following Ding et. al. (2023) and Ding et. al. (2024), distinct training codes have been provided for various datasets.

For simplicity, we only show how to implement the proposed CCDM in each experiment.

Before running the experiment, navigate to the ./CCDM/CCDM_unifed directory in your terminal.

(1) RC-49 (64x64)

To execute the training process, run the script ./scripts/RC64/win/run_ccdm.bat on Windows or ./scripts/RC64/linux/run_ccdm.sh on Linux. Make sure to correctly configure the following path: YOUR_PATH.