Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
executable file 147 lines (93 sloc) 5.22 KB

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:

Data download:

630 target images:

Fixation maps:

Fixation locations:

Annotation structure: allImages_release.mat

Displaying LabelMe annotations on top of images:


LabelMe Matlab toolbox


But run the LMplot.m provided in this directory instead of the one in the LabelMe toolbox.

>> im = imread(allImages(i).impath);
>> LMplot(allImages(i).annotation,im);

Using the allImages structure:

After downloading the, to display an image:

>> imshow(allImages(i).impath)

Computing the memorability score (HR) using the MTurk (AMT) data:

>> a = allImages(i).AMT_comb;
>> HR = a.hits/(a.hits+a.misses)

The encoding fixations of the j-th participant on the i-th image:

>> allImages(i).userdata(j).fixations.enc

Note: each image was shown to participants 3 times, denoted: encoding (enc), recognition (rec), second recognition (rec2). If recorded successfully for a particular participant, fixations may be available for any or all of these 3 times.

The (keypress) responses of the j-th participant on the i-th image:

>> allImages(i).userdata(j).SDT

Note: SDT is a 3-element vector corresponding to the recorded responses ('signal detections') of the participant on the encoding, recognition, and second recognition image repetitions.

The SDT values are as follows: 1 = hit; 2 = false alarm; 3 = miss; 4 = correct rejection

So, the j-th participant recognized the i-th image on the second presentation if:

>> allImages(i).userdata(j).SDT(2) == 1

Note: the j-th participant is the same for all images, so if participant j did not see image i during the experimental setting, allImages(i).userdata(j) will have empty fields.

Recomputing fixation maps:

To recreate fixation maps and fixation locations maps per image using the allImages structure:

(1) Download:



code for visualizations

(2) Run the generate_fixMaps.m file provided here

Retrieving the original images and annotations:

Example image: airport_terminal/sun_aabkzjntjfarengi.jpg

Then the corresponding SUN database folder is: a/airport_terminal (the first letter of the category name, followed by the category name itself)

The original full-sized image can be found at:

The LabelMe annotations for this image can be found at:

Example to download images and annotations from LabelMe and save them into a struct:

>> folderlist = {'users/antonio/static_sun_database/a/airport_terminal','users/antonio/static_sun_database/b/badlands'};
>> D = LMdatabase(HOMEANNOTATIONS,HOMEIMAGES,folderlist);

Hint: The LabelMe toolbox function 'LMimresizecrop' can be used to resize the image and annotations jointly. (see also 'LMvalidobjects' and 'addcroplabel')

Note: the following pairs of images are duplicates:

Targets/highway/sun_bnjxztsvnercygxn.jpg ~ Targets/highway/sun_bdwttbytrbnqyqsk.jpg Targets/conference_room/sun_awjucfvtjllxrtkv.jpg ~ Targets/conference_room/sun_abpqxslcljhrwmck.jpg Targets/highway/sun_byoonkqxghujgvkh.jpg ~ Targets/highway/sun_bjitfqyiepkgfkks.jpg Targets/highway/sun_bwrlzawyknljbhcb.jpg ~ Targets/highway/sun_auwrraazjwdcjcjg.jpg Targets/mountain/sun_boskvwzgsvsyuhll.jpg ~ Targets/mountain/sun_bgdykfpjgudqpzlu.jpg Targets/conference_room/sun_bzagimnxconenicn.jpg ~ Targets/conference_room/sun_bsccnfecifucnavf.jpg