Face Recognition Vendor Test
The Face Recognition Vendor Test is a standardized way of analyzing the performance of facial recognition software. This test was created to measure how accurately facial recognition systems can detect gender, age, ethnicity, and emotion. Due to the large number of tests available for vendors to take at no cost, it is becoming increasingly difficult for Buyers and Vendors to accurately evaluate vendor performance without manually conducting tests. The goals are to provide a standard test that will be easily accessible by both Buyers and Vendors so that when they engage in future discussions about selection decisions, they can do so on a level playing field with accurate statistics and data.
Try out free Face Recognition Vendor Test! https://www.facialverificationtest.com https://www.facialverificationtest.com/fvtv/
This is the first version of this free test. Due to the high quality facial recognition technologies that will be released in 2017, we are excited to bring you this test so you can see through your competitors' software and better determine the right vendor for you! Vendor Test information will not be shared with 3rd parties or sold for purposes other than testing performance. Vendors who take the test can opt-in to receive anonymized statistics about their own performance as well as aggregated results pertaining to their industry when they sign up for a FREE account on https://www.facialverificationtest.com .
Face Recognition Vendor Test is currently conducted on a per-vendor basis. In this project, we have created the Facial Recognition Vendor Test Template which can be found under the Resources section of our website. This document will be available for all vendors to use as a template when participating in the project, but we are interested in hearing your feedback! Please feel free to share any suggestions or ideas, as well as any concerns or wishes you may have, by contacting us at info@facialverificationtest.com .
The Face Recognition Vendor Test (FRVT) is a series of face detection and recognition benchmarks. In this test the subjects are upright frontal human faces, captured with a fixed resolution of 72 dots per inch. The images representing the test subjects were selected from four different databases: LFPW, Yale, BU, and CASIA. The FRVT 2012 has implemented three evaluation scenarios as follows:
The final score for each vendor was calculated using a weighted average across all three tests: Averaged Precision (AP), Average Matching Score (AMS), and Average Mutual Information (AMI). All scores were normalized to lie in the range [0-1].
Comments
Post a Comment