Explore

 

Dataset

Our dataset is comprised of images from various retargeting papers, and augmented by new ones, to create a diverse set of image types and content. The retargeting results were created using both state of the art methods as well as simple ones such as uniform scaling and manual cropping.

All images and results

Retargeting operators (see the supplemental for more details)

Objective measures (see the supplemental for more details)

Image attributes

 

User Study

Check out our live survey system for a demo!

Survey Design - design decisions and snapshots.

Instructions page - first page in the survey. Contains general instructions and motivation.

Comparison page - the main comparison interface. In this example, a comparison is made between the SV (left) and WARP (right) results on the boat image. Of course the user is unaware of the operators which generated the two results, and the side in which each result is shown is randomly chosen when the page is rendered.

Question page - such question was given to participants sporadically throughout the survey. Assuming the user has favored the WARP result in the previous comparison, this is the page that would be displayed. The list of options is composed according to the attributes that were marked for that specific image. Other images might have different options according to their attributes.

No-reference (blind) version

This version of the survey was intended to check whether user resizing preferences change when unaware of the original content. The survey pages were only slightly modified (the reference image was removed from the trial pages as well), and overlap of participants between the two versions was avoided. The question page was not applicable for this version of the survey.

Comparison page - the comparison page is shown here for the CR (left) and SM (right) operators on the Canalhouse image. Notice how the interface helps emphasizing the differences between the two results, that would otherwise be harder to grasp if the images were presented side by side.

 

Analysis

Subjective Analysis

Users' responses analysis - analysis of the data collected in the study

User votes by image - reference version

User votes by image - no-reference (blind) version

Question analysis

Objective Analysis

Metric analysis

Metric results by image - all metric distances

Rank comparison by image - objective and subjective rankings per image

Correlation distribution - distribution of the ranking correlation overall and by attribute

Miscellaneous

User statistics

Performance statistics

User comments