- BiRefNet General
- Open weights · commercially permissiveLicense: MIT
- BiRefNet HRSOD
- Open weights · commercially permissiveLicense: MIT
birefnet-general vs birefnet-hrsod
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-hrsod |
|---|---|---|
| MGE (Edge Quality) | 0.04117 ✓ | 0.05566 |
| MAE (Overall Accuracy) | 0.01632 ✓ | 0.03030 |
- MGE
- 0.03911✓
- MAE
- 0.00742✓
- MGE
- 0.10666
- MAE
- 0.03756
- MGE
- 0.05151✓
- MAE
- 0.02941✓
- MGE
- 0.08378
- MAE
- 0.12069
- MGE
- 0.03935✓
- MAE
- 0.00627✓
- MGE
- 0.05336
- MAE
- 0.00989
- MGE
- 0.01441✓
- MAE
- 0.00777✓
- MGE
- 0.02760
- MAE
- 0.01410
- MGE
- 0.02280✓
- MAE
- 0.00639✓
- MGE
- 0.03619
- MAE
- 0.01169
- MGE
- 0.05248
- MAE
- 0.01589✓
- MGE
- 0.04905✓
- MAE
- 0.01791
- MGE
- 0.03986
- MAE
- 0.01194
- MGE
- 0.01962✓
- MAE
- 0.00544✓
- MGE
- 0.06897✓
- MAE
- 0.01163✓
- MGE
- 0.09296
- MAE
- 0.01683
- MGE
- 0.02118✓
- MAE
- 0.00394✓
- MGE
- 0.02913
- MAE
- 0.00554
- MGE
- 0.03558✓
- MAE
- 0.01109✓
- MGE
- 0.03945
- MAE
- 0.01697
- MGE
- 0.04585✓
- MAE
- 0.00903✓
- MGE
- 0.05330
- MAE
- 0.01138
- MGE
- 0.04642✓
- MAE
- 0.00943✓
- MGE
- 0.06630
- MAE
- 0.01412
- MGE
- 0.02089
- MAE
- 0.00573
- MGE
- 0.01989✓
- MAE
- 0.00551✓
- MGE
- 0.03183✓
- MAE
- 0.00786✓
- MGE
- 0.04531
- MAE
- 0.01049
- MGE
- 0.05580✓
- MAE
- 0.01033✓
- MGE
- 0.14535
- MAE
- 0.05671
- MGE
- 0.01899✓
- MAE
- 0.00728✓
- MGE
- 0.03191
- MAE
- 0.01630
- MGE
- 0.01820✓
- MAE
- 0.01234✓
- MGE
- 0.02268
- MAE
- 0.01404
- MGE
- 0.08594✓
- MAE
- 0.02513✓
- MGE
- 0.13722
- MAE
- 0.11387
- MGE
- 0.01814✓
- MAE
- 0.00291✓
- MGE
- 0.02933
- MAE
- 0.00535
- MGE
- 0.02291✓
- MAE
- 0.00565✓
- MGE
- 0.04154
- MAE
- 0.00969
- MGE
- 0.00578✓
- MAE
- 0.00237✓
- MGE
- 0.01345
- MAE
- 0.00483
- MGE
- 0.01608✓
- MAE
- 0.00625✓
- MGE
- 0.03087
- MAE
- 0.01259
- MGE
- 0.03751✓
- MAE
- 0.00610✓
- MGE
- 0.05176
- MAE
- 0.00892
- MGE
- 0.03350
- MAE
- 0.02219✓
- MGE
- 0.02781✓
- MAE
- 0.02580
- MGE
- 0.01732✓
- MAE
- 0.00341✓
- MGE
- 0.02812
- MAE
- 0.00556
- MGE
- 0.00992✓
- MAE
- 0.00353✓
- MGE
- 0.01795
- MAE
- 0.00599
- MGE
- 0.01705✓
- MAE
- 0.00698✓
- MGE
- 0.02646
- MAE
- 0.00951
- MGE
- 0.05592✓
- MAE
- 0.07340✓
- MGE
- 0.09121
- MAE
- 0.23094
- MGE
- 0.11893
- MAE
- 0.06936
- MGE
- 0.07902✓
- MAE
- 0.06336✓
- MGE
- 0.13372
- MAE
- 0.09454
- MGE
- 0.12241✓
- MAE
- 0.04154✓
- MGE
- 0.09324✓
- MAE
- 0.01624✓
- MGE
- 0.13557
- MAE
- 0.02569
- MGE
- 0.03565✓
- MAE
- 0.01114✓
- MGE
- 0.04528
- MAE
- 0.02559
- MGE
- 0.05451✓
- MAE
- 0.01951✓
- MGE
- 0.06522
- MAE
- 0.03561
- MGE
- 0.02037✓
- MAE
- 0.01244✓
- MGE
- 0.02651
- MAE
- 0.02019
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).