mirror of
https://github.com/deepinsight/insightface.git
synced 2026-05-18 06:38:19 +00:00
InsightFace
Face Recognition Project
Experiments
Default image size is 112x96 if not specified, all face images are aligned.
In ResNet setting, _v1 means original residual units. _v2 means pre-activation units. _v3 means BCBACB residual units.
_bo means using bottleneck residual units.
- Softmax on LFW
| Network/Dataset | VGG2@112x112 | WebFace | MS1M | - | - |
|---|---|---|---|---|---|
| SE-ResNet50_v3 | 0.9973 | - | - | ||
| SE-ResNet50_v3_bo | |||||
| SE-ResNet50_v1 | |||||
| ResNet50_v3 | - | - | - | ||
| Inception-ResNet | - | - | - | ||
| SE-Inception-ResNet | - | - | - | ||
| MobileNet | - | - | - | ||
| ResNeXt | - | - | - | ||