- U²-Net
- Open weights · commercially permissiveLicense: Apache-2.0
- U²-Net Human Segmentation
- Open weights · commercially permissiveLicense: Apache-2.0
u2net vs u2net_human_seg
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 | u2net | u2net_human_seg |
|---|---|---|
| MGE (Edge Quality) | 0.07730 ✓ | 0.08689 |
| MAE (Overall Accuracy) | 0.06708 ✓ | 0.09628 |
- MGE
- 0.14337✓
- MAE
- 0.07828✓
- MGE
- 0.18039
- MAE
- 0.26527
- MGE
- 0.06854✓
- MAE
- 0.22669✓
- MGE
- 0.06967
- MAE
- 0.22743
- MGE
- 0.09070✓
- MAE
- 0.03029✓
- MGE
- 0.09961
- MAE
- 0.03528
- MGE
- 0.03809✓
- MAE
- 0.01951✓
- MGE
- 0.04501
- MAE
- 0.02340
- MGE
- 0.06467✓
- MAE
- 0.01425✓
- MGE
- 0.08194
- MAE
- 0.06321
- MGE
- 0.06502
- MAE
- 0.03815
- MGE
- 0.06375✓
- MAE
- 0.03799✓
- MGE
- 0.04867✓
- MAE
- 0.03375✓
- MGE
- 0.05262
- MAE
- 0.06527
- MGE
- 0.17699✓
- MAE
- 0.06924✓
- MGE
- 0.19185
- MAE
- 0.24021
- MGE
- 0.07183✓
- MAE
- 0.14568
- MGE
- 0.08905
- MAE
- 0.11216✓
- MGE
- 0.04437✓
- MAE
- 0.02576✓
- MGE
- 0.05164
- MAE
- 0.02787
- MGE
- 0.07678✓
- MAE
- 0.05786✓
- MGE
- 0.09263
- MAE
- 0.05819
- MGE
- 0.09232✓
- MAE
- 0.02617✓
- MGE
- 0.10398
- MAE
- 0.03386
- MGE
- 0.02496✓
- MAE
- 0.00703✓
- MGE
- 0.03566
- MAE
- 0.01279
- MGE
- 0.07380✓
- MAE
- 0.03234
- MGE
- 0.07784
- MAE
- 0.02957✓
- MGE
- 0.17586
- MAE
- 0.09050✓
- MGE
- 0.16755✓
- MAE
- 0.20554
- MGE
- 0.04105✓
- MAE
- 0.02754✓
- MGE
- 0.05299
- MAE
- 0.03977
- MGE
- 0.03046✓
- MAE
- 0.02603✓
- MGE
- 0.04393
- MAE
- 0.03770
- MGE
- 0.15255✓
- MAE
- 0.29758✓
- MGE
- 0.17383
- MAE
- 0.37089
- MGE
- 0.08677
- MAE
- 0.07768
- MGE
- 0.07626✓
- MAE
- 0.05722✓
- MGE
- 0.08256✓
- MAE
- 0.03681✓
- MGE
- 0.08966
- MAE
- 0.32712
- MGE
- 0.02242
- MAE
- 0.01491
- MGE
- 0.02200✓
- MAE
- 0.01058✓
- MGE
- 0.03497✓
- MAE
- 0.01392✓
- MGE
- 0.04587
- MAE
- 0.02406
- MGE
- 0.11639✓
- MAE
- 0.14357✓
- MGE
- 0.12973
- MAE
- 0.23347
- MGE
- 0.03938✓
- MAE
- 0.04264✓
- MGE
- 0.04527
- MAE
- 0.04463
- MGE
- 0.03781✓
- MAE
- 0.00835✓
- MGE
- 0.04340
- MAE
- 0.00965
- MGE
- 0.02301✓
- MAE
- 0.00770✓
- MGE
- 0.03144
- MAE
- 0.01333
- MGE
- 0.03719✓
- MAE
- 0.02523✓
- MGE
- 0.04789
- MAE
- 0.03824
- MGE
- 0.05676✓
- MAE
- 0.03121✓
- MGE
- 0.06552
- MAE
- 0.03158
- MGE
- 0.08390✓
- MAE
- 0.11076✓
- MGE
- 0.09699
- MAE
- 0.11154
- MGE
- 0.12890✓
- MAE
- 0.05244✓
- MGE
- 0.15004
- MAE
- 0.13050
- MGE
- 0.23568✓
- MAE
- 0.11443✓
- MGE
- 0.26149
- MAE
- 0.19606
- MGE
- 0.05487✓
- MAE
- 0.07367
- MGE
- 0.05666
- MAE
- 0.04061✓
- MGE
- 0.07571✓
- MAE
- 0.24833
- MGE
- 0.07874
- MAE
- 0.08652✓
- MGE
- 0.03168✓
- MAE
- 0.03248
- MGE
- 0.03927
- MAE
- 0.03209✓
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).