How kentucky-kaneki/adversarial_chest_xray_models works
This repository is a chest X-ray adversarial-robustness experiment, centered on a notebook and three DenseNet-121 variants. Its documented workflow compares a baseline, a clean-image fine-tuned model, and an adversarially trained robust model under several attacks, including a CLAHE comparison. The tree also contains model checkpoints, sample arrays, and result tables. Source excerpts were unavailable, so the graph shows only README-supported relationships; links from the notebook to individual result files are omitted rather than assumed.
Main entry points: densenet121_robust.pth Robust model · adversarial_xray_combined.ipynb Adversarial attacks · baseline Baseline model · densenet121_finetuned.pth Fine-tuned model
Components 13
Experiment
- Experiment notebook
home/adversarial_xray_combined.ipynb - Sample arrays
home/sample_images.npy
Model variants
- Baseline model
home/results/baseline - Fine-tuned model
home/working/densenet121_finetuned.pth - Robust model
home/working/densenet121_robust.pth
Evaluation outputs
- Adversarial attacks
home/adversarial_xray_combined.ipynb - CLAHE comparison
home/adversarial_xray_combined.ipynb - Baseline results
home/results/baseline - Fine-tuned results
home/results/fine-tuned - Robust results
home/results/robust - Comparative analyses
home/results/others
Other components
- Researcher
- NIH Chest X-ray
Connections 8
- Researcher → Experiment notebook: executes
README.md - NIH Chest X-ray → Baseline model: trains
README.md - NIH Chest X-ray → Fine-tuned model: trains
README.md - NIH Chest X-ray → Robust model: trains
README.md - Baseline model → Adversarial attacks: evaluated under
README.md - Fine-tuned model → Adversarial attacks: evaluated under
README.md - Robust model → Adversarial attacks: evaluated under
README.md - Robust model → CLAHE comparison: compared with
README.md