# Accuracy benchmark — 2026-05-15T20:28:15Z

| field | value |
|---|---|
| date | `2026-05-15T20:28:15Z` |
| py_feat_version | `0.7.0` |
| git_sha | `67cd7d9` |
| host | `liquidswords2` |
| gpu | `NVIDIA RTX PRO 6000 Blackwell Workstation Edition` |
| torch | `2.11.0+cu128` |
| python | `3.12.13` |
| device | `cuda` |
| omp_num_threads | `1` |

**Detector config**

| stage | model |
|---|---|
| face_model | `img2pose` |
| landmark_model | `mobilefacenet` |
| au_model | `xgb` |
| emotion_model | `resmasknet` |
| facepose_model | `img2pose` |
| identity_model | `arcface` |

## DISFA+ (posed peak) — AU intensity

- samples: **4500**, faces detected: 4500
- elapsed: 306.21s (14.7 fps)
- **AU F1 mean: 0.3607**
- **AU ICC mean: 0.1359**

| AU | F1 | ICC |
|---|---|---|
| AU01 | 0.4640 | 0.1315 |
| AU02 | 0.5124 | 0.1330 |
| AU04 | 0.4821 | 0.1430 |
| AU05 | 0.0540 | 0.0349 |
| AU06 | 0.5215 | 0.2099 |
| AU09 | 0.2037 | 0.0758 |
| AU12 | 0.5378 | 0.2813 |
| AU15 | 0.0979 | 0.0361 |
| AU17 | 0.1333 | 0.0184 |
| AU20 | 0.1023 | 0.0495 |
| AU25 | 0.6717 | 0.3304 |
| AU26 | 0.5470 | 0.1864 |