This is a simple implementation of VLAD in pure PyTorch (with kornia).
In bash
python -m pytorch_vlad --train-dir /PATH/TO/ROOT/IMAGE/DIR
or in python
import torch
import pytorch_vlad as vlad
model = vlad.train(**kwargs)
index_df, index_df_path = vlad.index(**kwargs)
retr_indices, retr_df = vlad.retrieve(**kwargs)
TorchVLAD
is distributed under the terms of the MIT license.