ModelAuditor Pre-trained Models

Pre-trained models for medical image classification, used with the ModelAuditor framework for AI-powered model auditing and robustness evaluation.

Models

Model	Architecture	Domain	Task	Input Size
camelyon17_resnet50_1_224.pt	ResNet50	Pathology	Tumor detection in lymph node sections	224x224
chexpert_resnet50_1_224.pt	ResNet50	Radiology	Chest X-ray classification	224x224
ham10000_resnet50_1_224.pt	ResNet50	Dermatology	Skin lesion classification (melanoma vs. benign keratosis)	224x224
cifar10.pth	ResNet50	General	CIFAR-10 image classification	224x224
DermaMNIST_resnet18.pth	ResNet18	Dermatology	Skin lesion classification (7 classes)	224x224

Usage

Download Models

pip install huggingface_hub

# Download all models
huggingface-cli download lukaskuhndkfz/ModelAuditor --local-dir models

# Or download individually
huggingface-cli download lukaskuhndkfz/ModelAuditor ham10000_resnet50_1_224.pt --local-dir models

Use with ModelAuditor

git clone https://github.com/lukaskuhndkfz/ModelAuditor
cd ModelAuditor
pip install -e ".[medical]"

# Run auditing
python main.py --model resnet50 --dataset ham10000 --weights models/ham10000_resnet50_1_224.pt

Load in PyTorch

import torch
from torchvision.models import resnet50

model = resnet50(num_classes=2)
model.load_state_dict(torch.load("ham10000_resnet50_1_224.pt", map_location="cpu"))
model.eval()

For DermaMNIST (ResNet18):

import torch
from torchvision.models import resnet18

model = resnet18(num_classes=7)
model.load_state_dict(torch.load("DermaMNIST_resnet18.pth", map_location="cpu"))
model.eval()

Training

Training scripts for all ResNet50 models are available in this repository as well (click on Files and Versions in the menu above).

Datasets

• Camelyon17: https://wilds.stanford.edu/datasets/#camelyon17
• CheXpert: https://stanfordmlgroup.github.io/competitions/chexpert/
• HAM10000: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/DBW86T
• CIFAR-10: https://www.cs.toronto.edu/~kriz/cifar.html
• DermaMNIST: https://medmnist.com/