FIGRIM Dataset

About

images

This is a dataset of 9428 images, 1754 of which are target images for which we obtained memorability scores.
The images span 21 scene categories from the SUN database. Each scene category was chosen to contain at least 300 images of size 700x700 or greater.
All images were cropped to 700x700 pixels.

experiments

We collected memorability scores using the experimental set-up of Isola et al. (CVPR 2011), by putting up memory games on Amazon's Mechanical Turk (AMT).
Images were shown to participants in sequence for 1 sec each (separated by a fixation cross). Targets were repeated once in the sequence.
Participants responded (using a keypress) anytime they detected an image had repeated in the sequence:


In AMT 1, we ran an experiment per scene category (for all 21 categories). Each target image was shown in the context of other images in the same category.
In AMT 2, we combined target images across all categories. Each target image was shown in the context of all other images.

memorability scores

We recorded a response as a hit if a participant correctly detected a repeated image.
A miss is when an image was repeated but the participant did not respond.
A false alarm is when an image shown for the first time was responded to.
A correct rejection is when an image was shown for the first time, and the participant did not make a keypress.
On average, about 80 participants saw each target image and its repeat.
The hits, misses, false alarms, and correct rejections are tallied across all participants, for each image. These are the scores we provide.
Using these scores, hit rate and false alarm rate for an image I can be computed as:

download

All target images come with 2 sets of memorability scores, corresponding to two sets of experiments:
where an image appears in the context of images of the same scene category (AMT 1)
and where an image appears in the context of images of all categories (AMT 2).
Filler images are provided only for context modeling (no scores are collected for filler images).

images

download file structure with all images (targets and fillers, sorted by category, 700x700 pixels)
OR
download matlab matrix with target images (matrix of all images, 200x200 pixels) [?]
download matlab matrix with filler images (matrix of all images, 200x200 pixels)

memorability scores

download csv file for within-category (AMT 1) scores
download csv file for across-category (AMT 2) scores
OR
download matlab data struct for target images (with within-category and across-category memorability scores) [?]
download matlab data struct for filler images (lists all filenames of filler images)

citation

@misc{figrim,
author = {Zoya Bylinskii and Phillip Isola and Constance Bainbridge
           and Antonio Torralba and Aude Oliva},
title        = {Intrinsic and Extrinsic Effects on Image Memorability},
howpublished = {Under review}
}