README-EN.md

Faiss Web Service

Overview

This project is forked from plippe/faiss-web-service .

Then I add three new feature:

1. Image feature extraction
2. Build Faiss index
3. Build docker which can run Faiss.

For more detail info about Faiss and this project, you can check my blog.

1. Image feature extraction

This project use SIFT of OpenCv, code location is: src/utils/feature_detect.py .

2. Build Faiss index

A Simple Way

Code location：src/train_index/train_index.py. This type index doesn't need train.

Build index which need pre-train

In the real world, we may use the index which need pre-train, such as "IVFx,Flat".

About how to choose what type index you need, you can refer here.

Code location：src/train_index/train_index_with_pre_train.py.

BoW

If you think the output dimension of SIFT (default is 128) is too low, you can use the Bag of Words. Code location：src/train_index/train_index_bow.py。

Build index from Javacv output feature

Code location：src/train_index/train_index_from_java.py。

Begin

Prepare

I have install Opencv 3.2 for extract feature base on original docker image. You can go the docker hub image download base image:

docker pull waltyou/faiss-api-service:1.2.1-gpu

API Using Guide

Builde Index

Start docker container：

cd src/train_index
python train_index.py

Search API

# Faiss search for ids 1, 2, and 3
curl 'localhost:5000/faiss/search' -X POST -d '{"k": 5, "ids": [1, 2, 3]}'

# Faiss search for image path
curl 'localhost:5000/faiss/search' -X POST -d '{"k": 5, "image": “/image/path/imagename”}'

# Faiss search for vector file path
curl 'localhost:5000/faiss/search' -X POST -d '{"k": 5, "vectors": “/vector/file/path”}'

Run

Go into docker container , and run bin/faiss_web_service.sh .

Check Status

Check if container is ok：

# Healthcheck
curl 'localhost:5000/ping'