An ML inference server to serve Image classification over HTTP
Vishal Keshav
Kenneth Myers
Jessie Huo
Following script starts the docker
sh start_docker_script.sh
Following script does the inference on a sample image if the app is running in the docker
sh infer_image_script.sh
A built docker image can be downloaded from this public link: https://umass.box.com/s/8p67ps7zpjwi6t3lyxbzdx67qgg1lp4p
docker, docker-compose, curl
On linux machine, use the following commands (uses apt)
sudo apt install docker
sudo apt install docker-compose [Install this for one step docker setup]
sudo apt install curl
From the repository main directory,
Step1: sudo docker-compose up
This will start the server at 0.0.0.0:5000
For developer's, if you change anything in the source, please run sudo docker-compose build before sudo docker-compose up. This re-builds the docker with changes in the source file.
Once the server is on, you can either use UI or command line interface.
Use curl command to query inference on the local images. The generic command is shown below:
curl -X POST -F "file=@</path/image.jpeg>" http://0.0.0.0:5000/uploadajax where </path/image.jpeg> is the image file location.
Example: If the image file "cat.jpg" is located at the root of the file system, then the command will be:
curl -X POST -F "[email protected]" http://0.0.0.0:5000/uploadajax
Or if the image is located in the upload/ directory then you can use:
curl -X POST -F "file=@upload/cat.jpg" http://0.0.0.0:5000/uploadajax
The app can be accessed via http://0.0.0.0:5000/ and images can be uploaded by clicking on or dragging images to the drop zone. To test a new image simply click on the whitespace around the thumbnail (see image below). There are also some example images you can use to test the classifier.
The documentation can be found in the docs folder.