---
license: other
task_categories:
- image-classification
- medical-imaging
pretty_name: Opth Project 1 DR Online Adaptation Backup
---

# Opth Project 1

This repository backs up the datasets, model checkpoints, experiment outputs, figures, manifests, and code snapshot used for Project 1: clinician-confirmed online adaptation for retinal diabetic retinopathy classification.

## Contents

- `datasets/`: archived fundus datasets used in the experiments.
- `annotations/`: IDRiD lesion/structure annotations for future lesion-click and attention-guided adaptation experiments.
- `manifests/`: labels and source/target split metadata.
- `models/`: trained ResNet50 checkpoints and validation metadata.
- `results/`: online adaptation summaries and bootstrap outputs.
- `figures/`: generated paper tables and plots.
- `configs/`: label mapping and source-target matrix.
- `code/`: OAIMS/classification code snapshot used for the experiments.
- `docs/`: study notes, paper planning logs, and dataset notes.

## Primary Task

Referable diabetic retinopathy detection from fundus images, with grade 0/1 mapped to non-referable and grade 2/3/4 mapped to referable.

## Intended Use

This backup is intended for research reproducibility and migration of the project across machines. The dataset archives are provided with attribution to the original data sources. Users are responsible for complying with each source dataset's license and terms.

## Restore

Example:

```bash
tar -xf datasets/aptos2019.tar -C ./restored_datasets/
mkdir -p ./restored_datasets/eyepacs_augmented_resized_V2
tar -xf datasets/eyepacs_augmented_resized_v2_clean.tar -C ./restored_datasets/eyepacs_augmented_resized_V2/
```

See `dataset_inventory.tsv`, `manifest_summary.tsv`, and `checksums.sha256` for file inventory and verification.