Model Performance Metrics

True Positive Rate at specified False Positive Rate: IJB-C Dataset

False Positive Rate	TFV5	TFV6	TFV7
1 in 1,000	0.97827	0.97893	0.98359
1 in 10,000	0.96733	0.96993	0.977704
1 in 100,000	0.94657	0.95245	0.96661
1 in 1,000,000	0.87943	0.88592	0.86833
1 in 10,000,000	0.32669	0.40594	0.34913

Receiver Operating Characteristic (ROC) Tables

Use the following ROC curves and tables to choose the optimal operating threshold for your use case. Consult the attached blog post for instructions on how to read and use a ROC curve:

Trueface Tutorials: Understanding ROC Curves

The Receiver Operating Characteristics (ROC) curve is an evaluation metric for a binary classifier, which helps us to visualize the performance of a facial recognition model as its discrimination threshold changes. In binary classification, we make a binary decision using a continuous probability value of a given sample belonging to one of the two classes.

medium.com

Trueface Tutorials: Understanding ROC Curves

Below we provide operating similarity-score thresholds based on evaluations run on various datasets. Choose thresholds pertaining to the dataset which is most representative of your deployment data.

IJB-C

You can learn about the characteristics of the IJB-C dataset here: https://noblis.org/wp-content/uploads/2018/03/icb2018.pdf

Similarity Score Threshold at False Positive Rates

False Positive Rate	TFV5	TFV6	TFV7
1 in 1,000	0.26996	0.26210	0.22404
1 in 10,000	0.34768	0.33785	0.29976
1 in 100,000	0.44167	0.41902	0.38314
1 in 1,000,000	0.58387	0.57696	0.61790
1 in 10,000,000	0.85119	0.82764	0.85776

Mugshot Dataset [deprecated]

The following ROC curves were generating using a private mugshot dataset. The dataset primarily consists of male faces

TFV6

TAR at FPR 10^-5: 0.9923

(the title below is incorrect, should say TFV6).

TFV5

TAR at FPR 10^-5: 0.9900

TFV4

TAR at FPR 10^-5: 0.9797

LITE V2

TAR at FPR 10^-5: 0.9032

LITE

TAR at FPR 10^-5: 0.8487