Data for "Beyond Overconfidence: Model Advances and Domain Shifts Redefine Calibration in Neural Networks"

The repository contains model outputs from our study on neural network calibration across various architectures and domains.

Available Datasets:

• ImageNet
• ImageNet-V2
• ImageNet-A
• Breast Ultrasound
• Retinal OCT
• Pneumonia
• Derma (HAM10000)
• Derma-C

Each .safetensor file includes:

• Model logits and ground truth labels for the validation set (for the distribution-shifted datasets)
• Model logits and ground truth labels for the test set

File Naming Convention:

[model_name]_output.safetensor, where model_name is the string identifier from the timm library.

Available Models:

• ConvNeXt
• EVA
• BEiT
• ViT
• Swin Transformer
• ResNet

Note:

Due to size constraints, we have not been able to upload the ImageNet-C model outputs. However, these outputs can be made available upon request.