Painting retrieval

The aim of the project is to provide a telegram bot that can recognize a photographed painting and that can recommend similar ones.

Preview

Install

virtualenv venv
source venv/bin/activate

pip install -r requirements.txt
pip install -e .

Create an .env file in the root folder and insert your bot's token

TOKEN = 110201543:AAHdqTcvCH1vGWJxfSeofSAs0K5PALDsaw

Configuration

Download all_data_info.csv, train.zip and test.zip from this dataset and put them into painting_retrieval/data/raw/dataset/.

Unzip model/resnet_model.zip in the same directory.

Then, from the root, run the following script

pip install -e "/PATH_TO_PAINTING_RETREIVAL/painting_retrieval"

Now, from /PATH_TO_PAINTING_RETREIVAL/painting_retrieval/, run

python3 scripts/generate_dataframe.py
python3 scripts/extract_features.py 
python3 scripts/index_features.py

Execution

Run the bot

python3 main.py

Run any scripts

python3 scripts/{script-name}.py

Folders

data

This folder is intended to contain the original and processed dataset.

model

This folder contains all the trained models.

• resnet_model.zip : ResNet50 model finetuned.
• KMeans_BOW.joblib : KMeans model, used in BOW features
• Scaler_BOW.joblib : Scaler, used in BOW features

scripts

This folder contains the code that must be executed offline to set up the system before execution.

src

painting

This folder contains the code concerning the paintings. It ranges from the calculation of the features to the matching.

bot

This folder contains the code concerning the telegram bot.