This repository contains the code for our winning solution to the 2020 edition of the Google Landmark Recognition competition hosted on Kaggle:
The full solution is described in a paper hosted on arxiv:
In order to run this code you need the train and test data from GLDv2:
To train a model, please run
src/train.pywith a config file as flag:
python train.py --config config1
You need to adjust data paths and other parameters in respective config file to make it work.
The blending and ranking procedure is detailed in