- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- U²-Net
- Open weights · commercially permissiveLicense: Apache-2.0
BiRefNet General vs u2net
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 | u2net |
|---|---|---|
| MGE (Edge Quality) | 0.03384 ✓ | 0.05404 |
| MAE (Overall Accuracy) | 0.01691 ✓ | 0.06673 |



- MGE
- 0.03450✓
- MAE
- 0.00754✓
- MGE
- 0.09509
- MAE
- 0.08100



- MGE
- 0.03391✓
- MAE
- 0.02906✓
- MGE
- 0.03862
- MAE
- 0.22676



- MGE
- 0.03369✓
- MAE
- 0.00657✓
- MGE
- 0.06007
- MAE
- 0.03098



- MGE
- 0.01534✓
- MAE
- 0.00932✓
- MGE
- 0.03274
- MAE
- 0.01732



- MGE
- 0.01776✓
- MAE
- 0.00662✓
- MGE
- 0.04736
- MAE
- 0.01496



- MGE
- 0.05518✓
- MAE
- 0.02445✓
- MGE
- 0.06842
- MAE
- 0.04546



- MGE
- 0.02070✓
- MAE
- 0.00960✓
- MGE
- 0.03513
- MAE
- 0.03698



- MGE
- 0.06499✓
- MAE
- 0.01274✓
- MGE
- 0.11366
- MAE
- 0.07027



- MGE
- 0.01420✓
- MAE
- 0.00333✓
- MGE
- 0.04141
- MAE
- 0.14397



- MGE
- 0.03582✓
- MAE
- 0.01162✓
- MGE
- 0.04762
- MAE
- 0.02104



- MGE
- 0.02967✓
- MAE
- 0.00716✓
- MGE
- 0.06526
- MAE
- 0.02066



- MGE
- 0.01780✓
- MAE
- 0.00511✓
- MGE
- 0.02158
- MAE
- 0.00659



- MGE
- 0.02221✓
- MAE
- 0.00778✓
- MGE
- 0.03629
- MAE
- 0.03255



- MGE
- 0.04873✓
- MAE
- 0.01122✓
- MGE
- 0.09893
- MAE
- 0.09053



- MGE
- 0.01818✓
- MAE
- 0.00827✓
- MGE
- 0.03523
- MAE
- 0.02607



- MGE
- 0.01969✓
- MAE
- 0.01374✓
- MGE
- 0.02820
- MAE
- 0.02370



- MGE
- 0.05570✓
- MAE
- 0.02558✓
- MGE
- 0.07998
- MAE
- 0.29806



- MGE
- 0.01467✓
- MAE
- 0.00332✓
- MGE
- 0.04300
- MAE
- 0.07894



- MGE
- 0.02090✓
- MAE
- 0.00612✓
- MGE
- 0.06347
- MAE
- 0.03625



- MGE
- 0.00619✓
- MAE
- 0.00277✓
- MGE
- 0.02008
- MAE
- 0.01485



- MGE
- 0.01810✓
- MAE
- 0.00714✓
- MGE
- 0.03800
- MAE
- 0.01532



- MGE
- 0.03517✓
- MAE
- 0.00648✓
- MGE
- 0.06015
- MAE
- 0.14384



- MGE
- 0.02895✓
- MAE
- 0.02783✓
- MGE
- 0.03059
- MAE
- 0.04455



- MGE
- 0.01442✓
- MAE
- 0.00315✓
- MGE
- 0.03554
- MAE
- 0.00845



- MGE
- 0.01079✓
- MAE
- 0.00372✓
- MGE
- 0.02402
- MAE
- 0.00790



- MGE
- 0.01879✓
- MAE
- 0.00851✓
- MGE
- 0.03196
- MAE
- 0.02429



- MGE
- 0.04204✓
- MAE
- 0.07583
- MGE
- 0.04327
- MAE
- 0.02803✓



- MGE
- 0.09392
- MAE
- 0.06015✓
- MGE
- 0.06458✓
- MAE
- 0.10452



- MGE
- 0.07979
- MAE
- 0.09417
- MGE
- 0.07450✓
- MAE
- 0.05224✓



- MGE
- 0.09319✓
- MAE
- 0.01831✓
- MGE
- 0.16225
- MAE
- 0.11610



- MGE
- 0.03339✓
- MAE
- 0.01185✓
- MGE
- 0.04612
- MAE
- 0.07316



- MGE
- 0.04978✓
- MAE
- 0.01930✓
- MGE
- 0.07274
- MAE
- 0.23800



- MGE
- 0.01854✓
- MAE
- 0.00954✓
- MGE
- 0.02746
- MAE
- 0.02864
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).