A visual search engine based on Elasticsearch and Tensorflow

Install requirements

$ cd visual_search
$ pip install -r requirements.txt

Setup

• Setup Elasticsearch

The easiest way to setup is using Docker with Docker Compose. With docker-compose everything you have to do is so simple:

cd visual_search/elasticsearch
docker-compose up -d

• Install elasticsearch plugin

We need to build Elasticsearch plugin to compute distance between feature vectors. Make sure that you have Maven installed.

$ cd visual_search/es-plugin
$ mvn install

$ cd target/release
// create simple server to serve plugin
$ python -m 'SimpleHTTPServer' &

//install plugin
//go back to elasticsearch folder
$ cd ../../..
$ docker exec -it elasticsearch_elasticsearch_1 elasticsearch-plugin install http://localhost:8000/esplugin-0.0.1.zip
$ docker-compose restart

• Index preparation

curl -XPUT http://localhost:9200/im_data -d @schema_es.json

• Setup faster r-cnn

I used earlier faster r-cnn version implemented by @Endernewton for object detection. To get pretrained model, please visit release section, download and extract file model.tar.gz to visual_search/models folder.

You also need to build faster r-cnn library by running following commands:

$ cd visual_search/lib
$ make

Indexing images to elasticsearch

To index data, just run command:

$ python index_es.py --input [image dir]

For full comamnd options, please run:

$ python index_es.py --help

Start server

Before starting the server, you must to update IMGS_PATH variable in visual_search/server.py to the location of folder where images are stored.

cd visual_search
python server.py

Now, you can access the link http://localhost:5000/static/index.html to test the search engine.

LICENSE

MIT