Moffett AI offers a breakthrough Visual Search as a Service Solution architecture based on Sparse Processing on FPGAs for Face Identification and Visual Search which is widely used in surveillance, smart retail, social media and autonomous vehicles.
Moffett AI FPGA Accelerated Visual Search Engine and Service are ready made to accelerate Visual Search inference across a wide variety of industries and applications, such as surveillance, smart retail, content search, social media and autonomous vehicles, delivering an order of magnitude higher efficiency via reduced computational complexity and smaller network size, delivering up to 15 times lower cost compared to GPU based solutions.
This application is containerized and can be easily run in a few minutes in the cloud, or on-premises.
Follow the instructions based on your deployment method.
Login to the Xilinx Acceleration Application Store.
Create Access Key and Download it as “cred.json” File.
You can either run a single command or launch an interactive shell where you can run many commands.
Add storage as default
Add tags as wish
Configure Security Group. You create new one or use one of default. Set the "type" is "SSH" and "source" is "anywhere"
Review and launch
Select an existing key pair or create a new key pair then save it as pem file for ssh suck like "aws.pem".
The "FPGA Development AMI" requires 8 vCPUs. You may need request AWS to increase your account limit from here https://console.aws.amazon.com/support/home?region=us-east-1#/case/create and choose "Service limit increase".
SSH to a jump point server: xcoengvm232180, xcoengvm232197, xsjengvm210134 or xsjengvm210190. Then "ssh -i aws.pem centos@[aws_instance_ip]". You can find ip from the right.
git clone https://github.com/MoffettSystem/Xilinx_Base_Runtime.git /home/centos/Xilinx_Base_Runtime sudo -s source /home/centos/Xilinx_Base_Runtime/utilities/docker_install.sh sudo systemctl start docker sudo usermod -aG docker $USER
AWS_FPGA_REPO_DIR=/home/centos/aws-fpga git clone https://github.com/aws/aws-fpga.git $AWS_FPGA_REPO_DIR source $AWS_FPGA_REPO_DIR/sdk_setup.sh
Upload cred.jon file (get from step 1) to AWS instance in folder
/opt/xilinx/cred.json. Then chmod 777 cred.json.
Execute following commands:
git clone https://github.com/MoffettSystem/aws_visual_search.git /home/centos/aws_visual_search source /home/centos/aws_visual_search/utilities/moffett_run_docker_container.sh
The app will go through all the images in the query folder:
data/vsearch_sdk_demo/user_image/queries and it will return top 5 results for each query image.
Each result contains a pair of number including a matching image database index and its confidence score.