Ranking Server

Embeds the text queries into repsective spaces and computes scores for given queries.

This is part of the SOMHunter project. But is not dependant on in in any way since it is used by other tools as well (e.g. CVHunter). Therefore you must provide data and confgure this server separately.

Data & Configuration

First create a new folder called clip_data and add the binary file with frame features in it. Don't forget to update the filename in .env file as well.

Build & Run with Docker (recommended)

docker build -t ranking-server .
docker run  --rm -ti -p 8083:8083 -v ${PWD}:/ranking-server ranking-server sh run.sh

Build & Run

sh install.sh
sh run.sh

How to Query the Server

curl -v http://localhost:5354/clip/my%20text%20query

• returns float32 vector of size 640 representing the query in feature space

curl -v http://localhost:5354/clip-results/my%20text%20query

• returns int32 vector of size 10000 of sorted indexes of the most similar frames