A visual search engine based on Elasticsearch and Tensorflow

Requirements

There are serveral python libraries you must to install before building the search engine.

• elasticsearch==5.2.0
• Tensorflow==1.12.1
• Flask
• opencv-python

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

• Building 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
cd ../elasticsearch
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/img_data -d @schema_es.json

• Setup faster r-cnn

I use faster r-cnn version implemented by @Endernewton for object detection. Follow the link for the guide how to setup faster r-cnn and download VGG16 model trained on COCO dataset.

Indexing images to elasticsearch

export WEIGHT_PATH=...
export MODEL_PATH=...
export INPUT=..
cd visual_search
python index_es.py --weight $WEIGHT_PATH --model_path $MODEL_PATH --input $INPUT

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

and access the link http://localhost:5000/static/index.html to test the search engine.

Have fun =))