System Requirement
There are certain hardware and software-based requirements to deploy ID Analyzer On-Premise System with optimal performance.
CPU & Memory Requirements
OCR Module
To extract data from document images, an OCR engine must also be within the data extraction process, apart from ID Fort's proprietary A.I. document recognition system. ID Fort supports several different OCR engines with different performance and hardware requirements. You may set up the OCR engine you wish to use through the Administrator portal; if more than one OCR engine is selected, the document will be passed to all selected engines, and data with the highest confidence score will be used.
Built-In OCR Engine 1
| Requirement | Specification |
|---|---|
| Minimum Hardware | 4vCPU + 16GB RAM |
| Recommended Hardware | 8vCPU + GPU + 32GB RAM |
| Average Inference Time (Minimum Hardware) | 5s |
| Average Inference Time (Recommended Hardware) | 2s |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
| Overall Accuracy | 80% |
Built-In OCR Engine 2
| Requirement | Specification |
|---|---|
| Minimum Hardware | 4vCPU + 16GB RAM |
| Recommended Hardware | 8vCPU + GPU + 32GB RAM |
| Average Inference Time (Minimum Hardware) | 3s |
| Average Inference Time (Recommended Hardware) | 0.5-1.5s |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
| Overall Accuracy | 85% |
Azure OCR On-Premise (Read 3.2 Docker Container)
| Requirement | Specification |
|---|---|
| Minimum Hardware | 8vCPU + 16GB RAM |
| Recommended Hardware | 8vCPU + 24GB RAM |
| Average Inference Time (Minimum Hardware) | 1-1.5s |
| Average Inference Time (Recommended Hardware) | 1-1.5s |
| Requires Internet Connection | Yes, for billing purpose only |
| Extra Fee Required | Yes, customer responsible for related fees |
| Private Data Never Leave Premise | Yes |
| Overall Accuracy | 95% |
Azure Cloud OCR
| Requirement | Specification |
|---|---|
| Minimum Hardware | None |
| Average Inference Time (Minimum Hardware) | 0.5-1.5s |
| Requires Internet Connection | Yes |
| Extra Fee Required | Yes, customer responsible for related fees |
| Private Data Never Leave Premise | No, data will be sent to Microsoft |
| Overall Accuracy | 85% |
Baidu Cloud OCR
| Requirement | Specification |
|---|---|
| Minimum Hardware | None |
| Average Inference Time (Minimum Hardware) | 3-5s |
| Requires Internet Connection | Yes |
| Extra Fee Required | Yes, customer responsible to setup Azure billing |
| Private Data Never Leave Premise | No, data will be sent to Baidu |
| Overall Accuracy | 85% |
Tencent Cloud OCR
| Requirement | Specification |
|---|---|
| Minimum Hardware | None |
| Average Inference Time (Minimum Hardware) | 1-3s |
| Requires Internet Connection | Yes |
| Extra Fee Required | Yes, customer responsible for related fees |
| Private Data Never Leave Premise | No, data will be sent to Tencent |
| Overall Accuracy | 92% |
Choice of OCR engines depends on several factors:
- Accuracy Requirements
- Response Time Requirements
- Hardware Capabilities
- Data Privacy Concerns
For example, if you do not have a GPU server, we recommend using Azure OCR On-Premise, which relies on CPU AVX2 for fast inference. You could use any cloud service provider for faster inference and reduced server hardware requirements.
Biometric Module
Similar to OCR engines, ID Fort comes with the built-in biometric engine, which have low hardware requirements, however, it is possible to use Azure's face recognition engine with slightly higher accuracy.
Biometric Engine - 1:1 Face Verification
| Requirement | Specification |
|---|---|
| Minimum Hardware | 2vCPU + 4GB RAM |
| Recommended Hardware | 4vCPU + 8GB RAM |
| Average Inference Time (Minimum Hardware) | 0.5s |
| Average Inference Time (Recommended Hardware) | 0.3s |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
| Overall Accuracy | 98% |
Biometric Engine - Face Liveness Verification
| Requirement | Specification |
|---|---|
| Minimum Hardware | 4vCPU + 8GB RAM |
| Recommended Hardware | 4vCPU + GPU + 8GB RAM |
| Average Inference Time (Minimum Hardware) | 1.5 s / Photo or 12s / Video |
| Average Inference Time (Recommended Hardware) | 0.3s / Photo or 2.5s / Video |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
| Overall Accuracy | 96% |
Anti-Forgery Module
ID Fort's advanced anti-forgery engine can run on both CPU and GPU.
| Requirement | Specification |
|---|---|
| Minimum Hardware | 4vCPU + 16GB RAM |
| Recommended Hardware | 8vCPU + GPU + 16GB RAM |
| Average Inference Time (Minimum Hardware) | 2-3.5s |
| Average Inference Time (Recommended Hardware) | 0.5-1.5s |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
AML Module
ID Fort's advanced anti-forgery engine can run on both CPU and GPU.
| Requirement | Specification |
|---|---|
| Minimum Hardware | 2vCPU + 2GB RAM |
| Recommended Hardware | 2vCPU + 2GB RAM |
| Average Inference Time (Minimum Hardware) | 0.1s |
| Average Inference Time (Recommended Hardware) | 0.1s |
| Requires Internet Connection | No |
| Extra Fee Required | No |
| Private Data Never Leave Premise | Yes |
Benchmarks
We have performed benchmark tests under the following settings:
- AML + Biometric + Anti-forgery modules are enabled
- Input document image size of 1200x800
- Input face image size of 800x600
- Running either cloud OCR or built-in OCR, or Azure On-premise OCR
Using the table below, you can calculate the instances required to support your transaction volume per second.
| Server Specification | Built-in OCR Engines | Azure On-Premise | Cloud-based OCR |
|---|---|---|---|
| 4 vCPU + 16GB RAM | 6s / Transaction Max 0.25 TPS | 5s / TransactionMax 0.4 TPS | 3s / TransactionMax 1 TPS |
| 4 vCPU + T4 GPU +16GB RAM | 4s / Transaction Max 0.7 TPS | 5s / TransactionMax 0.4 TPS | 3s / TransactionMax 1 TPS |
| 8 vCPU + 16GB RAM | 4.5s / Transaction Max 0.6 TPS | 4s / TransactionMax 0.8 TPS | 2s / TransactionMax 3 TPS |
| 8 vCPU + T4 GPU +16GB RAM | 3s / Transaction Max 1 TPS | 4s / TransactionMax 0.8 TPS | 2s / TransactionMax 3 TPS |
| 16 vCPU + 32GB RAM | 3s / Transaction Max 1.2 TPS | 3s / TransactionMax 1.8 TPS | 1.5s / TransactionMax 5 TPS |
| 16 vCPU + T4 GPU + 32GB RAM | 2s / Transaction Max 2 TPS | 3s / TransactionMax 1.8 TPS | 1.5s / TransactionMax 5 TPS |
Storage Requirements
ID Fort Docker image requires a minimum 10 GB space to deploy. Transaction data can be stored locally inside the container or a remote MySQL database, depending on whether ID Fort runs in single or multi-instance mode. Transaction images can also be stored locally inside the container, in a remote MySQL database or on AWS S3.
The following disk spaces are required. If both transaction data and document images are stored within the container or inside a MySQL database:
- Transaction Data + 1500px Front Side Document = 250KB per transaction
- Transaction Data + 1000px Front Side Document = 130KB per transaction
- Transaction Data + 750px Front Side Document = 90KB per transaction
- Transaction Data + 1500px Front Side Document + 500px Face (biometric verification) = 310KB per transaction
- Transaction Data + 1000px Front Side Document + 500px Face (biometric verification) = 190KB per transaction
- Transaction Data + 750px Front Side Document + 500px Face (biometric verification) = 150KB per transaction
- Transaction Data + 1500px Front Side Document + 1500px Back Side Document + 500px Face (biometric verification) = 530KB per transaction
- Transaction Data + 1000px Front Side Document + 1000px Back Side Document + 500px Face (biometric verification) = 290KB per transaction
- Transaction Data + 750px Front Side Document + 750px Back Side Document + 500px Face (biometric verification) = 210KB per transaction
The following disk spaces are required. If transactions are stored within the container or inside a MySQL database, but the transaction images are stored on S3:
- Transaction Data = 30KB per transaction
Operating System Requirements
ID Fort can run on any OS that supports Docker Engine, which means you can run it on almost all Linux distributions and Windows as well. For ID Fort to utilize GPU on GPU-enabled servers, you must deploy the ID Fort container on an operating system that supports NVIDIA Container Toolkit - https://github.com/NVIDIA/nvidia-docker
Other Requirements
ID Fort API and web portal run through HTTPS, therefore, it is recommended to obtain a valid SSL certificate to secure the connection. If no SSL certificate is supplied, ID Fort can automatically obtain free certificates from Let's Encrypt if a valid hostname is set.
Production Architecture
To deploy ID Fort for production, it is recommended to:
- Carefully calculate the anticipated transaction volume and deploy a corresponding number of instances that is capable of handling busy-hour TPS volume.
- Deploy a minimum of two instances to prevent unanticipated downtime.
- Using a load balancer to balance the load between multiple ID Fort instances.
- MySQL database should be set up so that it is automatically replicated and backed up, an automatic failover is recommended.
Updated 9 months ago