DiffExplainer: Towards cross-modal global explanations with diffusion models

Matteo Pennisi, Giovanni Bellitto, Simone Palazzo, Isaak Kavasidis, Mubarak Shah, Concetto Spampinato

Quickstart

Environment

• Install uv if you don't have it.
• Create the environment and install deps:

  uv venv --python 3.9
  source .venv/bin/activate
  uv pip install -r requirements.txt

Models

• For --cnn resnet50, no extra files are needed.
• For --cnn robust50, download robust_resnet50.pth from the Salient ImageNet repo: https://github.com/singlasahil14/salient_imagenet/ and place it in the project root.

Run

python main.py --cnn resnet50 --class_to_activate 537 --optim_steps 1000 --diff_steps 4 --device cuda:0

Adjust --feature_idx to include the feature loss term, or omit it for CE-only.

Citation

If you use this repository in your research, please cite it as:

@article{PENNISI2025104559,
title = {DiffExplainer: Towards cross-modal global explanations with diffusion models},
journal = {Computer Vision and Image Understanding},
volume = {262},
pages = {104559},
year = {2025},
issn = {1077-3142},
doi = {https://doi.org/10.1016/j.cviu.2025.104559},
url = {https://www.sciencedirect.com/science/article/pii/S1077314225002826},
author = {Matteo Pennisi and Giovanni Bellitto and Simone Palazzo and Isaak Kavasidis and Mubarak Shah and Concetto Spampinato},
}