---
language:
- en
license: mit
task_categories:
- image-to-text
size_categories:
- 100K<n<1M
---
# Image Reconstructions for CyclePrefDB-I2T

[Project page](https://cyclereward.github.io) | [Paper](https://huggingface.co/papers/2506.02095) | [Code](https://github.com/hjbahng/cyclereward)

This dataset contains reconstruction images used to determine cycle consistency preferences for CyclePrefDB-I2T. You can find the corresponding file paths in the CyclePrefDB-I2T dataset [here](https://huggingface.co/datasets/carolineec/CyclePrefDB-I2T). Reconstructions are created using [Stable Diffusion 3 Medium](https://huggingface.co/stabilityai/stable-diffusion-3-medium).

## Preparing the reconstructions

You can download the test and validation split `.tar` files and extract them directly.

Use this script to extract the training files:

```
cat train_recons.tar.gz.* | tar xzvf -
```