%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% COPYRIGHT %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% This package contains Matlab source code for visualizing quantization effects for bag-of-visual-words representationsi as described in: Alexander Freytag and Johannes Ruehle and Paul Bodesheim and Erik Rodner and Joachim Denzler: "Seeing through bag-of-visual-word glasses: towards understanding quantization effects in feature extraction methods". ICPR Workshop on Features and Structures (FEAST), 2014 Please cite that paper if you are using this code! (LGPL) copyright by Alexander Freytag and Johannes Ruehle and Paul Bodesheim and Erik Rodner and Joachim Denzler %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% START / SETUP %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % adapt this file according to your system, including path to 3party projects. % Mirrors for downloading libraries not yet existing in your systems are displayed. initWorkspace % NOTE - additional data needed Calling inverseHog for the first time will download the paired-ditionary file, which is around 15MB. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DEMO 1 Invert a given image, following the BoW paradigm %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %run the whole *inversion* pipeline, including local feature extraction, codebook generation, inversion, and impainting out = demo1_inversionPipeline % visualize the resulting image figure; imshow ( out.imgHoggleBow ) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DEMO 2 Invert a given image and randomly shuffle its parts %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %run the whole pipeline, including local feature extraction, codebook generation, inversion, and impainting out = demo2_imageInvertedAndJittered % visualize the resulting image figure; imshow ( out.imgHoggleBow ) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DEMO 3 Evaluate a standard BoW-pipeline %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % NOTE - dataset needed % This demo uses the 15 scenes dataset. You need to adapt data/init15Scenes.m according to the location of the 15scenes dataset on your machine. % The dataset is avaible at http://www.cs.illinois.edu/homes/slazebni/research/scene_categories.zip % NOTE - setup 15Scenes dataset % after downloading the dataset and extracing it to /HOME/MYHOME/MYFOLDER15SCENES/ % go to folder of dataset cd /HOME/MYHOME/MYFOLDER15SCENES/ % run the script provided with this repo to write all image filenames into class-specific filelists /HOME/MYHOME/MYCODE/bowClassification.git/data/list_files.sh jpg 15Scenes.txt filelist % copy file with references to the filelists into the data directory (alternative, change the path to this file in the demo3_script) cp /HOME/MYHOME/MYFOLDER15SCENES/15Scenes.txt /HOME/MYHOME/MYCODE/bowClassification.git/data/15Scenes.txt %run the whole *classification* pipeline, including local feature extraction, codebook generation, quantization, training, and testing out = demo3_evaluateBoWPerformance % NOTE - small subset only % per default, demo3 only runs on a subset of the 15scenes dataset. You might want to change the settings in order to evaluate the pipeline on the whole dataset, or even a completely different one.
cvjena/bowInversion
Visualizing quantization effects in common bag-of-visual-words representations fro images.
MATLABLGPL-3.0