in the wild
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Audio Deepfake Detection | In-the-wild | EER0.8 | 76 | |
| Image Forgery Localization | In the Wild | DSC63.3 | 15 | |
| Image Forgery Localization | In the Wild Unseen Domain | mIoU51 | 15 | |
| Novel View Synthesis | In-the-wild data | PSNR29.26 | 14 | |
| Illumination Estimation | in-the-wild Samsung | Mean Angular Error2.07 | 13 | |
| Illumination Estimation | in-the-wild Pixel | Mean Angular Error (°)1.89 | 13 | |
| Face retouching | In-the-wild 1,000 internet portraits (test) | NIQE11.821 | 11 | |
| Pixel-level Forgery Localization | In-the-wild | F1 Score69.18 | 11 | |
| Image-level forgery detection | In-the-wild | F1 Score100 | 11 | |
| Voice Anti-spoofing | In-the-Wild (test) | EER6.24 | 11 | |
| Bounding Box Localization | In-the-wild | BBox IoU63.54 | 10 | |
| Fake Image Detection | In-the-Wild CommunityAI SocialRF | CommunityAI NLL0.7879 | 8 | |
| Ordinal consistency | In-the-wild 100 steps horizon v1 (test) | Kendall's Tau0.61 | 8 | |
| Ordinal consistency | In-the-wild 50 steps horizon v1 (test) | Kendall's Tau0.69 | 8 | |
| Reference-guided Image Super-Resolution | In-the-wild Customization | CLIP-I0.728 | 7 | |
| Reference-guided Image Super-Resolution | In-the-wild Compositing | CLIP-I0.8476 | 7 | |
| Audio Anti-spoofing | In-the-Wild | EER1.88 | 7 | |
| Face Reenactment | in the wild | AU %82.3 | 7 | |
| 3D Avatar Reconstruction | In-the-Wild (test) | L1 Loss0.008 | 6 | |
| Synthetic Speech Detection | In-The-Wild (ITW) | EER15.39 | 6 | |
| Novel View Synthesis | In-the-Wild Composite | PSNR26.94 | 6 | |
| Single-view 3D Generation | in-the-wild (test) | Uni3D Score42.11 | 5 | |
| Single-image head reconstruction | In-the-wild | ArcFace0.414 | 5 | |
| Dynamic Scene Generation | In-the-wild dataset | PhysReal0.82 | 5 | |
| Music-driven 2D dance generation | In-the-Wild leakage-free (test) | FID45.2 | 5 |
Showing 25 of 42 rows