- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- BiRefNet Portrait
- Open weights · commercially permissiveLicense: MIT
birefnet-general vs birefnet-portrait
Quantitative comparison: 33 images scored against ground-truth mattes. Lower is better. Use the view controls on each row to switch between transparency, chroma key, and alpha matte on all outputs.
| Metric | birefnet-general | birefnet-portrait |
|---|---|---|
| MGE (Edge Quality) | 0.03384 ✓ | 0.04243 |
| MAE (Overall Accuracy) | 0.01691 ✓ | 0.05090 |
- MGE
- 0.03450✓
- MAE
- 0.00754✓
- MGE
- 0.08386
- MAE
- 0.09753
- MGE
- 0.03391✓
- MAE
- 0.02906✓
- MGE
- 0.03811
- MAE
- 0.22692
- MGE
- 0.03369✓
- MAE
- 0.00657✓
- MGE
- 0.04008
- MAE
- 0.00974
- MGE
- 0.01534✓
- MAE
- 0.00932
- MGE
- 0.01560
- MAE
- 0.00698✓
- MGE
- 0.01776✓
- MAE
- 0.00662✓
- MGE
- 0.04045
- MAE
- 0.02887
- MGE
- 0.05518✓
- MAE
- 0.02445✓
- MGE
- 0.05656
- MAE
- 0.03090
- MGE
- 0.02070✓
- MAE
- 0.00960✓
- MGE
- 0.02176
- MAE
- 0.01088
- MGE
- 0.06499✓
- MAE
- 0.01274✓
- MGE
- 0.10133
- MAE
- 0.10167
- MGE
- 0.01420✓
- MAE
- 0.00333✓
- MGE
- 0.02262
- MAE
- 0.00455
- MGE
- 0.03582
- MAE
- 0.01162
- MGE
- 0.03076✓
- MAE
- 0.00979✓
- MGE
- 0.02967✓
- MAE
- 0.00716✓
- MGE
- 0.04891
- MAE
- 0.01535
- MGE
- 0.01780
- MAE
- 0.00511
- MGE
- 0.01050✓
- MAE
- 0.00324✓
- MGE
- 0.02221✓
- MAE
- 0.00778✓
- MGE
- 0.03766
- MAE
- 0.02335
- MGE
- 0.04873✓
- MAE
- 0.01122✓
- MGE
- 0.09709
- MAE
- 0.04888
- MGE
- 0.01818✓
- MAE
- 0.00827✓
- MGE
- 0.01919
- MAE
- 0.00875
- MGE
- 0.01969✓
- MAE
- 0.01374
- MGE
- 0.02099
- MAE
- 0.01246✓
- MGE
- 0.05570✓
- MAE
- 0.02558✓
- MGE
- 0.07146
- MAE
- 0.22746
- MGE
- 0.01467✓
- MAE
- 0.00332✓
- MGE
- 0.02543
- MAE
- 0.00925
- MGE
- 0.02090✓
- MAE
- 0.00612✓
- MGE
- 0.05887
- MAE
- 0.32245
- MGE
- 0.00619✓
- MAE
- 0.00277✓
- MGE
- 0.00747
- MAE
- 0.00291
- MGE
- 0.01810✓
- MAE
- 0.00714✓
- MGE
- 0.02263
- MAE
- 0.00842
- MGE
- 0.03517✓
- MAE
- 0.00648✓
- MGE
- 0.05296
- MAE
- 0.02716
- MGE
- 0.02895
- MAE
- 0.02783
- MGE
- 0.02091✓
- MAE
- 0.02130✓
- MGE
- 0.01442
- MAE
- 0.00315
- MGE
- 0.01334✓
- MAE
- 0.00231✓
- MGE
- 0.01079✓
- MAE
- 0.00372
- MGE
- 0.01116
- MAE
- 0.00319✓
- MGE
- 0.01879✓
- MAE
- 0.00851✓
- MGE
- 0.02111
- MAE
- 0.00942
- MGE
- 0.04204
- MAE
- 0.07583
- MGE
- 0.02939✓
- MAE
- 0.01078✓
- MGE
- 0.09392
- MAE
- 0.06015
- MGE
- 0.03867✓
- MAE
- 0.02312✓
- MGE
- 0.07979✓
- MAE
- 0.09417✓
- MGE
- 0.08326
- MAE
- 0.22116
- MGE
- 0.09319✓
- MAE
- 0.01831✓
- MGE
- 0.15441
- MAE
- 0.10118
- MGE
- 0.03339
- MAE
- 0.01185
- MGE
- 0.03066✓
- MAE
- 0.01006✓
- MGE
- 0.04978✓
- MAE
- 0.01930✓
- MGE
- 0.05210
- MAE
- 0.02417
- MGE
- 0.01854✓
- MAE
- 0.00954✓
- MGE
- 0.02091
- MAE
- 0.01539
Methodology
- Alpha mattes are grayscale masks encoding per-pixel foreground probability. They capture fine edge detail (hair, fur, semi-transparent regions) that binary masks discard.
- Metrics are computed by comparing predicted mattes against a ground-truth reference: MGE measures edge sharpness and MAE measures pixel-level accuracy. Both are lower-is-better.
- Scored images come from Bench v0, where every image has a ground-truth matte. The benchmark page has the full methodology and per-image scores for every model (CSV / JSON).