New baselines

Included two new baselines trained with ArcFace head These two were trained by myself using MSCeleb.

resnet50-msceleb-arcface-2021

=====================  ================  =================
..                     Development       Evaluation
=====================  ================  =================
Failure to Acquire     0.0%              0.0%
False Match Rate       0.5% (299/57960)  0.4% (531/147630)
False Non Match Rate   0.5% (13/2520)    1.5% (59/3990)
False Accept Rate      0.5%              0.4%
False Reject Rate      0.5%              1.5%
Half Total Error Rate  0.5%              0.9%
=====================  ================  =================

mobilenetv2-msceleb-arcface-2021

[Min. criterion: EER ] Threshold on Development set `/idiap/temp/tpereira/temp/mobilenetv2-mobio-male/scores-dev`: -9.859437e-01
=====================  ================  ==================
..                     Development       Evaluation
=====================  ================  ==================
Failure to Acquire     0.0%              0.0%
False Match Rate       1.1% (621/57960)  1.0% (1534/147630)
False Non Match Rate   1.1% (27/2520)    2.5% (100/3990)
False Accept Rate      1.1%              1.0%
False Reject Rate      1.1%              2.5%
Half Total Error Rate  1.1%              1.8%
=====================  ================  ==================

To use it do in a FR experiment do:

bob bio pipelines vanilla-biometrics <database> resnet50-msceleb-arcface-2021 ....

bob/bio/face/config/baseline/helpers.py

@@ -235,7 +235,11 @@ def embedding_transformer_112x112(
     cropped_image_size = (112, 112)
     if annotation_type == "eyes-center":
         # Hard coding eye positions for backward consistency
-        cropped_positions = {"leye": (49, 72), "reye": (49, 38)}
+        cropped_positions = {
+            "leye": (55, 81),
+            "reye": (55, 42),
+        }
+
     else:
         # Will use default
         cropped_positions = embedding_transformer_default_cropping(