HTTP Segment

This is a simple service that wraps the rembg library in an HTTP API based on the Starlette framework.

Usage

This app can be run with make run or with Docker:

docker build -t $IMAGE_NAME .
docker run -p 8000:8000 $IMAGE_NAME

By specifying the MODEL environment variable you can change the model loaded by the service. The default is silueta.

Check this list to see what models are available.

Configuration

The following environment variables should be used to configure the service:

S3_ACCESS_KEY
S3_SECRET_KEY
S3_BUCKET
S3_ENDPOINT_URL

Endpoints

`/segment`

Accepts a POST request with a an image in the body of the request.

It will then perform segmentation on the image, upload the segmented images to S3 and return a JSON response with the URLs of the uploaded images, including the original.

curl -X POST --data-binary @./.github/readme/test.png http://localhost:8000/segment

{"file_names":["209d43982bc349299f45234071d1dd3d.webp","eb4c8fe8636648b790741ff29351fbf6.webp"]}

`/segment/{resize:int}`

The same as /segment but will resize the cropped images to the specified size, upscaling if necessary.

curl -X POST --data-binary @./.github/readme/test.png http://localhost:8000/segment/256

{"file_names":["209d43982bc349299f45234071d1dd3d.webp","eb4c8fe8636648b790741ff29351fbf6.webp"]}

`/health`

Returns a 200 response if the service is healthy. This is a bit of a weird healthcheck since it will return a 500 if two minutes have passed since the last request. This can be used to determine if the service is still running but hasn't received any requests in a while.

Since the AI model is loaded into memory when the first request is received, rather than at startup, this endpoint can be used to restart the service and free memory after a period of inactivity.

I built this mechanism because I don't expect this service to be used very often and I want to keep memory in check.

curl http://localhost:8000/health

TommasoAmici / http-segmentation