Welcome to my implementation of the C++ Final Project. This readme outlines the steps needed to get the project up and running and also discusses certain nuances of the way things are structured.
Boost and OpenCV4 are the only two prerequisites. The respective installation instructions can be found here and here. Please make sure to also install OpenCV's extra modules.
The project was tested using OpenCV version 4.3 and Boost version 1.66.0.
Assuming you have already downloaded the latest artifacts, simply head to the directory named bin and look for the file called main. This is the only file you will need to run the project (with it's default settings).
The program takes a number of command line arguments, described as below, to set various parameters that tune the functions to your liking and also optionally allows you to simply fix certain parameters in a configuration file to make things a little less tedious.
General Options:
-h [ --help ] display help message
-v [ --verbose ] print verbose output
-c [ --config-file ] arg path to the configuration file
-I [ --image-path ] arg path to image dataset
-D [ --descriptor-path ] arg path to precomputed feature descriptors
-H [ --histogram-path ] arg path to precomputed image histograms
-Q [ --query-path ] arg path to query image(s)
Configuration Options:
--use-flann arg use FLANN for histogram computations
(default true)
--use-opencv-kmeans arg use opencv kmeans implementation
(default true)
-k [ --num-clusters ] arg number of clusters
(default 100)
-m [ --max-iter ] arg maximum number of iterations
(default 25)
-e [ --epsilon ] arg stop iterations if specified accuracy,
epsilon, is reached
(default 1e-6)
-n [ --num-similar ] arg number of similar images to find
(default 10)
--reweight arg perform TF-IDF reweighting for
histograms
(default false)
--save-histograms arg save histogram dataset to disk
(default true)
--save-descriptors arg save descriptors dataset to disk
(default false)
-Q [ --query-path ] arg path to query image(s)
All configuration options can also be specified on the command line, in which case the values specified on the command line will take preference over those in the configuration file.
A sample configuration file, named bow_params.cfg can also be found under the bin directory.
The program assumes a certain directory structure for the dataset that it uses to store/load files and complains otherwise.
<dataset_root_dir>
├── <any_name> # Directory where the (png) image dataset is stored
├── descriptors # Directory where the descriptor dataset is stored
└── histograms # Directory where the histogram dataset is stored
├── codebook # The computed codebook too is stored in this directory
└── idf # And so are the inverse document frequencies, if any
Note that the descriptor and histogram files are stored with the same name as the original image.