- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- BiRefNet Massive
- Open weights · commercially permissiveLicense: MIT
birefnet-general vs birefnet-massive
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-massive |
|---|---|---|
| MGE (Edge Quality) | 0.03384 ✓ | 0.03965 |
| MAE (Overall Accuracy) | 0.01691 ✓ | 0.02234 |
- MGE
- 0.03450✓
- MAE
- 0.00754✓
- MGE
- 0.05840
- MAE
- 0.03213
- MGE
- 0.03391✓
- MAE
- 0.02906✓
- MGE
- 0.03695
- MAE
- 0.22434
- MGE
- 0.03369
- MAE
- 0.00657
- MGE
- 0.03036✓
- MAE
- 0.00569✓
- MGE
- 0.01534✓
- MAE
- 0.00932
- MGE
- 0.01950
- MAE
- 0.00826✓
- MGE
- 0.01776
- MAE
- 0.00662
- MGE
- 0.01413✓
- MAE
- 0.00270✓
- MGE
- 0.05518
- MAE
- 0.02445
- MGE
- 0.05214✓
- MAE
- 0.02378✓
- MGE
- 0.02070✓
- MAE
- 0.00960✓
- MGE
- 0.02709
- MAE
- 0.01414
- MGE
- 0.06499
- MAE
- 0.01274
- MGE
- 0.06317✓
- MAE
- 0.01272✓
- MGE
- 0.01420✓
- MAE
- 0.00333✓
- MGE
- 0.01567
- MAE
- 0.00398
- MGE
- 0.03582
- MAE
- 0.01162
- MGE
- 0.03000✓
- MAE
- 0.00914✓
- MGE
- 0.02967✓
- MAE
- 0.00716✓
- MGE
- 0.03776
- MAE
- 0.01013
- MGE
- 0.01780✓
- MAE
- 0.00511✓
- MGE
- 0.01792
- MAE
- 0.00976
- MGE
- 0.02221
- MAE
- 0.00778
- MGE
- 0.01997✓
- MAE
- 0.00759✓
- MGE
- 0.04873✓
- MAE
- 0.01122✓
- MGE
- 0.05595
- MAE
- 0.01593
- MGE
- 0.01818✓
- MAE
- 0.00827✓
- MGE
- 0.04070
- MAE
- 0.02035
- MGE
- 0.01969✓
- MAE
- 0.01374✓
- MGE
- 0.02327
- MAE
- 0.01633
- MGE
- 0.05570✓
- MAE
- 0.02558✓
- MGE
- 0.06721
- MAE
- 0.03654
- MGE
- 0.01467✓
- MAE
- 0.00332✓
- MGE
- 0.01545
- MAE
- 0.00387
- MGE
- 0.02090✓
- MAE
- 0.00612
- MGE
- 0.02449
- MAE
- 0.00583✓
- MGE
- 0.00619✓
- MAE
- 0.00277✓
- MGE
- 0.00943
- MAE
- 0.00447
- MGE
- 0.01810✓
- MAE
- 0.00714✓
- MGE
- 0.05837
- MAE
- 0.04824
- MGE
- 0.03517✓
- MAE
- 0.00648✓
- MGE
- 0.03597
- MAE
- 0.00685
- MGE
- 0.02895✓
- MAE
- 0.02783✓
- MGE
- 0.03465
- MAE
- 0.03383
- MGE
- 0.01442✓
- MAE
- 0.00315✓
- MGE
- 0.01852
- MAE
- 0.00400
- MGE
- 0.01079✓
- MAE
- 0.00372✓
- MGE
- 0.01949
- MAE
- 0.00672
- MGE
- 0.01879✓
- MAE
- 0.00851✓
- MGE
- 0.04253
- MAE
- 0.01640
- MGE
- 0.04204
- MAE
- 0.07583
- MGE
- 0.03509✓
- MAE
- 0.01486✓
- MGE
- 0.09392✓
- MAE
- 0.06015
- MGE
- 0.09978
- MAE
- 0.04155✓
- MGE
- 0.07979
- MAE
- 0.09417
- MGE
- 0.06949✓
- MAE
- 0.01981✓
- MGE
- 0.09319
- MAE
- 0.01831
- MGE
- 0.08811✓
- MAE
- 0.01729✓
- MGE
- 0.03339✓
- MAE
- 0.01185✓
- MGE
- 0.04840
- MAE
- 0.01911
- MGE
- 0.04978✓
- MAE
- 0.01930✓
- MGE
- 0.07133
- MAE
- 0.02808
- MGE
- 0.01854✓
- MAE
- 0.00954✓
- MGE
- 0.02711
- MAE
- 0.01281
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).