Scripts to prepare a visual search process.

Run:

pip install -r requirements.txt

And install FAISS

Before to download the dataset, you have to request an access to the owner of the dataset to have the download link. Once you have the download link put it in the download_amazon_dataset.py script, and then run the following command line:

python download_amazon_dataset.py

To create the embedding generator model, run the following command line:

python create_saved_model.py

If you want to run your Docker image over a GPU, you have to install Nvidia for Docker and run this command line:

docker run -d --gpus all --name serving_base tensorflow/serving:latest-gpu

Otherwise run this one:

docker run -d --name serving_base tensorflow/serving:latest

Then, run these commands:

mkdir model
mv resnet18 model
mkdir model/resnet18/1
mv model/resnet18/variables model/resnet18/saved_model.pb model/resnet18/1
docker cp model/resnet18 serving_base:/models/resnet18
docker commit --change "ENV MODEL_NAME resnet18" \
    --change "ENV PATH $PATH:/usr/local/nvidia/bin" \ 
    --change "ENV LD_LIBRARY_PATH /usr/local/nvidia/lib64" serving_base <image-name>
docker kill serving_base
docker rm serving_base

Run:

python create_faiss_index.py

To run a visual search over your images and index, you can use the corresponding backend and frontend.