SOR

Same Object Retrieval project for Computer Vision @ Sapienza/Computer-Science M.Sc

Reference Paper: An Improved Faster R-CNN for Same Object Retrieval

Tool

Slides

Instructions to run

• Clone the repository and create the following directories

  git clone https://github.com/Projectdotpy/SOR.git
  cd SOR
  mkdir data             # data folder containing the dataset
  mkdir model            # model folder containing the weights of the nn
  mkdir dist/query       # will contain the imgs given as query by the client
  cd dist
  ln -s ../data data     # needed to display the result in the browser
  cd ..

• Download instre_monuments.zip at this link. This dataset is a subset of images from the INSTRE dataset having only monuments. Move it under the data folder.

• Run complex_data_transformation.sh which will also run create_csv.py

  ./complex_data_transformation.sh

• Train the Network with the notebook present at this link. This step will create some files in drive namely model_frcnn_vgg.hdf5, record.csv and model_vgg_config.pickle

• Place the model_frcnn_vgg.hdf5 and record.csv files under model

• Run ./convert_pickle.py to modify the model_vgg_config.pickle

    ./convert_pickle.py

• Run create_retrieval_db.py. This may take a while

  ./create_retrieval_db.py # generates two files under retrieval_db

• Launch server.py

  ./server.py

• Browse to http://127.0.0.1:5000/index.html

Team

Davide Spallaccini	Beatrice Bevilacqua	Anxhelo Xhebraj	Indro Spinelli