API7 FR Server User Guide - Seventh Sense AI

API7 FR Server User Guide

v1.1.0

Introduction

Thank you for downloading API7 FR Server. This guide will help you get started with using the API7 FR Server. To install API7, you need a computer running Ubuntu 18.04/20.04 LTS. This can either be a physical machine or a VM on AWS/Azure/GCP.

System Requirements (physical machine)

The following table lists the minimum and recommended specifications of machines running the software:

OS

CPU Clock speed

RAM Free Disk Space* Registration Speed** Search Speed****

Minimum

Recommended

Ubuntu 18.04/20.04 LTS Desktop or Server

Intel Core i5 5th gen or later Intel Core i7 8th gen or later

2.2 GHz, 2 cores, 4 threads 3.2 GHz, 6 cores, 12 threads

8GB

8GB

600MB

2-3 RPS*** 7-8 RPS

1GB

4-5 RPS 12-13 RPS

* In addition to its initial install size, the product consumes 106.5MB per 10,000 enrolled persons with 1 picture each ** Registration speed is the speed of registering a person into the facial recognition database *** RPS ? Requests per second **** Search speed is the speed of searching for a person in the database with one image

The performance of the API depends on the hardware you install it on. The above speeds are from i5 5200U and Asus VivoMini VC66-C i7 8700 Intel Core CPUs.

Cloud

We have tested the software on the following instance types for AWS and Azure:

AWS ? t2.xlarge Azure - Standard_B4ms

You should get 12-13 RPS of search speed on the above instances.

All reported speeds are measured locally and do not take into account network latency.

Accuracy

We have evaluated our models on benchmarks such as LFW (Labeled Faces in the Wild) and IJB-A, IJB-C. Our best models were able to achieve an accuracy of 99.73 on the Labeled Faces in the Wild and 0.027 FNMR at FMR = 0.01 for IJB-A. This record beat the result of the National University of Singapore and Panasonic Team. The result is shown below:

These results are for "in the wild" matching where matches are not always trying to match a frontally oriented face with another frontally oriented face. Frontal to frontal matching is a much simpler task. With this in mind, we offer you two options while performing a search. You can select FAST as your search mode, which optimizes for fast search over accuracy or ACCURATE which prioritizes accuracy. Our ACCURATE mode offers a state of the art matching algorithm for high security scenarios and where matches may be hard due to challenges in lighting conditions, occlusions, change in age etc. For most casual frontal to frontal matching, for example if you are building an application which asks users to take a photo of themselves for recognition, our FAST mode should be good enough. If you plan to use ACCURATE mode more than FAST, we recommend that you choose a capable CPU or a bigger instance in the cloud. However, if your planned search capacity is well under 5 RPS (Requests per second), you can search using ACCURATE mode on our recommended system specification as well. The table on the following page lists the accuracy and size of our different search modes.

Search Mode

Model Size

LFW* Accuracy

Use Case

FAST

47.2 MB

99.62%

Frontal to frontal matching of

constrained images (for example ISO/IEC

19794-5).

Typically used by face capturing terminals

ACCURATE

130.7 MB

99.73%

"In the wild" matching of persons

of interest

*

Installation

To install the software, please follow the steps below: 1. Go to our website ? 2. Pick your OS and register your details. 3. Check your email for a download link, server license file and license key 4. Click the download link to download the file 5. Unzip the file on your physical machine or VM 6. Place the server-license.lic file attached to your email in the unzipped folder as shown on the following page:

7. To install the product, open a command prompt and then run ./install.sh

Trying the API

If you installed the product on your own computer, the API documentation page is available at

If you installed the product on a VM, you will need to open port 7777 to access the page from outside the VM. In the mail you received while downloading the product, you were also provided with a license key. This license key serves as a means to authorize requests with the API. Please click the Authorize button on the documentation page and enter your license key as shown:

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download