MagicBrush
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Instructive image editing | MagicBrush (test) | CLIP Image0.911 | 20 | |
| Visual World Modelling | MagicBrush | GPT-4o Score8.14 | 18 | |
| Instruction-guided image editing | MagicBrush single-turn (test) | CLIP Similarity (Image)0.9332 | 13 | |
| Image Editing | MagicBrush Single-Turn | L1 Loss0.033 | 11 | |
| Description-guided Image Editing | MagicBrush Multi-turn (test) | L1 Loss0.0911 | 10 | |
| Forward-dynamics Prediction | MagicBrush AURORA-BENCH | GPT-4o Score6.71 | 9 | |
| Instruction-based image editing | MagicBrush | L1 Loss0.0641 | 9 | |
| instruction-based object addition | MagicBrush | CLIP Score0.9312 | 7 | |
| Image Editing | MagicBrush v1 (test) | CLIP Input Similarity0.898 | 7 | |
| Image Editing | MagicBrush Multi-Turn | L1 Loss0.035 | 7 | |
| Instruction-guided image editing | MagicBrush multi-turn (test) | CLIP-T0.313 | 7 | |
| Image Editing | MagicBrush (test) | Overall Score3.26 | 7 | |
| Differences Caption Generation | MagicBrush (test) | Main Difference Count50 | 7 | |
| Edit Inspectors Question Answering | MagicBrush (test) | Accuracy73.7 | 7 | |
| Instruction-guided image editing | MagicBrush | LPIPS-U0.0446 | 6 | |
| Image Editing | MagicBrush | CLIPim91.1 | 6 |
Showing 16 of 16 rows