Code to supplement the FIGRIM (FIne-GRained Image Memorability) project and dataset.
Matlab C++ C
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
toolboxes
util first commit Jan 28, 2016
.DS_Store
README.md additional files added Apr 6, 2016
amt2demo_filenames.mat
calcMI.m
calculateScores.m first upload of all code files Apr 6, 2016
context_correlations.m
evaluateImageDistinctiveness.m
evaluateImageDistinctiveness_standalone.m
getContext.m
plotAllCorrsWithMem.m first upload of all code files Apr 6, 2016
plotCatCorrWithMem.m first commit Jan 28, 2016
plotCorrWithMem.m first commit Jan 28, 2016
sceneTypes.mat additional files added Apr 6, 2016

README.md

This data is available as a supplement to the following paper. If you use this data, please cite:

Zoya Bylinskii, Phillip Isola, Constance Bainbridge, Antonio Torralba, Aude Oliva. "Intrinsic and Extrinsic Effects on Image Memorability", Vision Research (2015)

Project page: http://figrim.mit.edu


Stand-alone files for computing feature distinctiveness and memorability scores

evaluateImageDistinctiveness_standalone.m computes the distinctiveness of an image relative to an image collection. Requires the input of image features for target images (for which the distinctiveness scores are desired) and filler images (used to compute the probability density for the image collection). This function computes a kernel density over all the image features in the collection, and then measures the negative log likelihood (distinctiveness) of a given image's features under this density.

calculateScores.m computes 5 different memorability scores: HR (hit rate), FAR (false alarm rate), ACC (accuracy), DPRIME (d-prime), and MI (mutual information). The calculations are included here: http://figrim.mit.edu/supplemental.pdf.

Reproduce the plots presented in the paper

Download the data files available here: http://figrim.mit.edu

context_correlations.m The main function to run for distinctiveness computations and correlation with memorability scores, for the within-category (AMT 1) and across-category (AMT 2) experiments.

getContext.m retrieves the images used for each of AMT 1 and AMT 2 experiments to reproduce the context used in each experiment for distinctiveness calculations.

plotAllCorrsWithMem.m create plots of different memorability scores as functions of image distinctiveness and category entropy.