README.md

---
license: apache-2.0
---

We evaluate DAEFR on one synthetic dataset **CelebA-Test**, and two real-world datasets **LFW-Test** and **WIDER-Test**.

<table>
<tr>
    <th>Datasets</th>
    <th>Filename</th>
    <th>Short Description</th>
    <th>Source</th>
</tr>
<tr>
    <td>CelebA-Test (HQ)</td>
    <td>celeba_512_validation.zip</td>
    <td>3000 (HQ) ground truth images for evaluation</td>
    <td><a href="https://github.com/wzhouxiff/RestoreFormer">RestoreFormer</a></td>
</tr>
<tr>
    <td>CelebA-Test (LQ)</td>
    <td>self_celeba_512_v2.zip</td>
    <td>3000 (LQ) synthetic images for testing</td>
    <td>Ourselves</td>
</tr>  
<tr>
    <td>LFW-Test (LQ)</td>
    <td>lfw_cropped_faces.zip</td>
    <td>1711 real-world images for testing</td>
    <td><a href="https://github.com/TencentARC/VQFR">VQFR</a></td>
</tr>
<tr>
    <td>WIDER-Test (LQ)</td>
    <td>Wider-Test.zip</td>
    <td>970 real-world images for testing</td>
    <td><a href="https://shangchenzhou.com/projects/CodeFormer/">CodeFormer</a></td>
</tr>
</table>