CDI Reference Models for CBED

# CDI Reference Models for CBED

This repository contains pretrained reference models for Conditional Diffusion Inference (CDI) experiments on the CBED dataset for magnesium diboride (MgB₂). The models are provided as tar archives containing complete CDI model directories with saved configurations, network weights, optimizer states, and training histories.

## Files

- `reference_diffusion_model.tar` — CDI conditional diffusion model for   uncertainty-aware CBED parameter inference.
- `reference_efficientnet_b2.tar` — deterministic EfficientNet-B2 regression baseline.
- `reference_resnet_50.tar` — deterministic ResNet-50 regression baseline.

## Archive Contents

Each archive unpacks to one model directory:

```text
reference_diffusion_model/
reference_efficientnet_b2/
reference_resnet_50/
```

The model directories include the saved training configuration (`config.json`), PyTorch model weights (`net_*.pth`), optimizer and scheduler states when available, and training/validation history files.

## Related Resources

- CDI code repository: https://github.com/realtime-intelligence/cdi-uq
- CBED-ML test split: https://zenodo.org/records/18340683
- CBED-ML training split, Part 1: https://zenodo.org/records/18344486
- CBED-ML training split, Part 2: https://zenodo.org/records/18356685
- CBED-ML training split, Part 3: https://zenodo.org/records/18476248