xukechun / CBIR

CBIR

Introduction

• This is a project of image retrieval for course CVML2020 in ZJU/CAML2020 in Cambridge.
• complement three methods（BOF/VLAD/BOF_SVM/BOF_VGG） for image retrieval.

Installation

• Python3

• cuda10.1, pytorch1.2.0, torchvision

• You can quickly install/update these dependencies by running the following (replace pip with pip3 for Python 3):

  pip install pickle sqlite3 opencv-python sklearn numpy scipy PCV PIL matplotlib argparse glob heapq itertools tqdm joblib
  

Instructions

1. generating vocabulary dictionaries

• run bof.py to generate vocabulary dictionary for BOF
• run vlad.py to generate vocabulary dictionary for VLAD
• run bof_svm.py to generate vocabulary dictionary for BOF_SVM
• run bof_vgg.py to generate vocabulary dictionary for BOF_VGG

2. testing with different methods

   User need to choose the ID of tested image(BOF and VLAD within 500 while BOF_SVM within 300 if using my dataset), number of images to retrieve and method to use(0:BOF,1:VLAD,2:BOF_SVM,3:BOF_VGG)

   Then run the following:

   python Search_images.py -q [Index of a query image] -r [number of images to retrieve] -m [method to use]