prismformore / delf-pytorch

This is a Pytorch reimplementation of "Large-Scale Image Retrieval with Attentive Deep Local Features"
13 stars 4 forks source link

delf-pytorch

This is a Pytorch implementation of DELF that is based on official tensorflow implementation. This repository provides clean training codes on The Quick Draw Dataset.

According to the paper, training process is divided into two steps: 1. Train resnet backbone; 2. Train attention layers. You may choose this in settings.py

Dataset should be generated from Simplified Drawing files (.ndjson) of Quick Draw Dataset.

Simply starting training by:

python main.py -a train