- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- BiRefNet DIS
- Open weights · commercially permissiveLicense: MIT
birefnet-general vs birefnet-dis
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-dis |
|---|---|---|
| MGE (Edge Quality) | 0.03384 ✓ | 0.04146 |
| MAE (Overall Accuracy) | 0.01691 ✓ | 0.02280 |
- MGE
- 0.03450✓
- MAE
- 0.00754✓
- MGE
- 0.04550
- MAE
- 0.01642
- MGE
- 0.03391✓
- MAE
- 0.02906✓
- MGE
- 0.03604
- MAE
- 0.22217
- MGE
- 0.03369✓
- MAE
- 0.00657✓
- MGE
- 0.03746
- MAE
- 0.00662
- MGE
- 0.01534✓
- MAE
- 0.00932✓
- MGE
- 0.02634
- MAE
- 0.01056
- MGE
- 0.01776
- MAE
- 0.00662
- MGE
- 0.01599✓
- MAE
- 0.00271✓
- MGE
- 0.05518✓
- MAE
- 0.02445✓
- MGE
- 0.07483
- MAE
- 0.04380
- MGE
- 0.02070✓
- MAE
- 0.00960✓
- MGE
- 0.02998
- MAE
- 0.01103
- MGE
- 0.06499
- MAE
- 0.01274
- MGE
- 0.06026✓
- MAE
- 0.01152✓
- MGE
- 0.01420
- MAE
- 0.00333✓
- MGE
- 0.01367✓
- MAE
- 0.00334
- MGE
- 0.03582✓
- MAE
- 0.01162✓
- MGE
- 0.04528
- MAE
- 0.01673
- MGE
- 0.02967✓
- MAE
- 0.00716✓
- MGE
- 0.03743
- MAE
- 0.00948
- MGE
- 0.01780
- MAE
- 0.00511✓
- MGE
- 0.01761✓
- MAE
- 0.00529
- MGE
- 0.02221✓
- MAE
- 0.00778✓
- MGE
- 0.02320
- MAE
- 0.00779
- MGE
- 0.04873✓
- MAE
- 0.01122✓
- MGE
- 0.05630
- MAE
- 0.01530
- MGE
- 0.01818✓
- MAE
- 0.00827✓
- MGE
- 0.04128
- MAE
- 0.01730
- MGE
- 0.01969✓
- MAE
- 0.01374✓
- MGE
- 0.03483
- MAE
- 0.01725
- MGE
- 0.05570✓
- MAE
- 0.02558✓
- MGE
- 0.06595
- MAE
- 0.08214
- MGE
- 0.01467✓
- MAE
- 0.00332✓
- MGE
- 0.01894
- MAE
- 0.01032
- MGE
- 0.02090✓
- MAE
- 0.00612✓
- MGE
- 0.02501
- MAE
- 0.00841
- MGE
- 0.00619✓
- MAE
- 0.00277✓
- MGE
- 0.01158
- MAE
- 0.00369
- MGE
- 0.01810✓
- MAE
- 0.00714✓
- MGE
- 0.03759
- MAE
- 0.01204
- MGE
- 0.03517✓
- MAE
- 0.00648
- MGE
- 0.03629
- MAE
- 0.00646✓
- MGE
- 0.02895✓
- MAE
- 0.02783✓
- MGE
- 0.03905
- MAE
- 0.03521
- MGE
- 0.01442✓
- MAE
- 0.00315✓
- MGE
- 0.01781
- MAE
- 0.00352
- MGE
- 0.01079✓
- MAE
- 0.00372✓
- MGE
- 0.02130
- MAE
- 0.00597
- MGE
- 0.01879✓
- MAE
- 0.00851✓
- MGE
- 0.03999
- MAE
- 0.01427
- MGE
- 0.04204✓
- MAE
- 0.07583
- MGE
- 0.04338
- MAE
- 0.01665✓
- MGE
- 0.09392
- MAE
- 0.06015
- MGE
- 0.09150✓
- MAE
- 0.03600✓
- MGE
- 0.07979
- MAE
- 0.09417
- MGE
- 0.07932✓
- MAE
- 0.01938✓
- MGE
- 0.09319
- MAE
- 0.01831✓
- MGE
- 0.08536✓
- MAE
- 0.01978
- MGE
- 0.03339✓
- MAE
- 0.01185✓
- MGE
- 0.04881
- MAE
- 0.01871
- MGE
- 0.04978✓
- MAE
- 0.01930✓
- MGE
- 0.07559
- MAE
- 0.02870
- MGE
- 0.01854✓
- MAE
- 0.00954✓
- MGE
- 0.03457
- MAE
- 0.01386
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).