- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- BiRefNet Massive
- Open weights · commercially permissiveLicense: MIT
birefnet-general vs birefnet-massive
Quantitative comparison: 34 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) | 0.04117 ✓ | 0.04789 |
| MAE (Overall Accuracy) | 0.01632 ✓ | 0.02234 |
- MGE
- 0.03911✓
- MAE
- 0.00742✓
- MGE
- 0.08779
- MAE
- 0.03215
- MGE
- 0.05151✓
- MAE
- 0.02941✓
- MGE
- 0.06524
- MAE
- 0.22443
- MGE
- 0.03935
- MAE
- 0.00627
- MGE
- 0.03289✓
- MAE
- 0.00524✓
- MGE
- 0.01441✓
- MAE
- 0.00777✓
- MGE
- 0.02357
- MAE
- 0.00949
- MGE
- 0.02280
- MAE
- 0.00639
- MGE
- 0.01515✓
- MAE
- 0.00255✓
- MGE
- 0.05248
- MAE
- 0.01589
- MGE
- 0.04678✓
- MAE
- 0.01543✓
- MGE
- 0.03986
- MAE
- 0.01194
- MGE
- 0.02603✓
- MAE
- 0.00953✓
- MGE
- 0.06897
- MAE
- 0.01163
- MGE
- 0.06309✓
- MAE
- 0.01117✓
- MGE
- 0.02118✓
- MAE
- 0.00394✓
- MGE
- 0.02421
- MAE
- 0.00473
- MGE
- 0.03558✓
- MAE
- 0.01109✓
- MGE
- 0.03806
- MAE
- 0.01303
- MGE
- 0.04585✓
- MAE
- 0.00903✓
- MGE
- 0.04698
- MAE
- 0.01346
- MGE
- 0.04642✓
- MAE
- 0.00943✓
- MGE
- 0.04706
- MAE
- 0.00964
- MGE
- 0.02089✓
- MAE
- 0.00573✓
- MGE
- 0.02171
- MAE
- 0.01033
- MGE
- 0.03183
- MAE
- 0.00786
- MGE
- 0.02978✓
- MAE
- 0.00774✓
- MGE
- 0.05580✓
- MAE
- 0.01033✓
- MGE
- 0.06964
- MAE
- 0.01467
- MGE
- 0.01899✓
- MAE
- 0.00728✓
- MGE
- 0.04751
- MAE
- 0.02203
- MGE
- 0.01820✓
- MAE
- 0.01234✓
- MGE
- 0.02909
- MAE
- 0.01790
- MGE
- 0.08594✓
- MAE
- 0.02513✓
- MGE
- 0.09822
- MAE
- 0.03573
- MGE
- 0.01814
- MAE
- 0.00291✓
- MGE
- 0.01614✓
- MAE
- 0.00291
- MGE
- 0.02291✓
- MAE
- 0.00565✓
- MGE
- 0.03223
- MAE
- 0.00654
- MGE
- 0.00578✓
- MAE
- 0.00237✓
- MGE
- 0.01128
- MAE
- 0.00504
- MGE
- 0.01608✓
- MAE
- 0.00625✓
- MGE
- 0.06164
- MAE
- 0.04938
- MGE
- 0.03751✓
- MAE
- 0.00610✓
- MGE
- 0.03880
- MAE
- 0.00657
- MGE
- 0.03350✓
- MAE
- 0.02219✓
- MGE
- 0.04063
- MAE
- 0.03091
- MGE
- 0.01732✓
- MAE
- 0.00341✓
- MGE
- 0.02280
- MAE
- 0.00450
- MGE
- 0.00992✓
- MAE
- 0.00353✓
- MGE
- 0.01898
- MAE
- 0.00647
- MGE
- 0.01705✓
- MAE
- 0.00698✓
- MGE
- 0.05230
- MAE
- 0.01700
- MGE
- 0.05592
- MAE
- 0.07340
- MGE
- 0.04452✓
- MAE
- 0.01430✓
- MGE
- 0.11893✓
- MAE
- 0.06936
- MGE
- 0.12430
- MAE
- 0.04741✓
- MGE
- 0.13372
- MAE
- 0.09454
- MGE
- 0.09991✓
- MAE
- 0.01996✓
- MGE
- 0.09324
- MAE
- 0.01624
- MGE
- 0.07595✓
- MAE
- 0.01400✓
- MGE
- 0.03565✓
- MAE
- 0.01114✓
- MGE
- 0.05668
- MAE
- 0.01968
- MGE
- 0.05451✓
- MAE
- 0.01951✓
- MGE
- 0.08162
- MAE
- 0.03282
- MGE
- 0.02037✓
- MAE
- 0.01244✓
- MGE
- 0.03768
- MAE
- 0.02286
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).