Training and Validation Set

The training/validation set contains the following:

  • the images,
  • the pair list,

    the list of the photo pairs along with the preferences of each pair (data and its labels)

  • the series information,

    including scores obtained from Bradley-Terry model and rankings in each series

  • the raw crowdsourced comments,

    the reasons of human preferences for further research purpose. In the Results > comparisons page, you could view the comments of a series by clicking the series.

  • the demo code.

    the code for evaluating the performance of your methods in the validation set

Testing Set

The testing set is used for our benchmark, and contains the following:

  • the images,
  • the pair list,

    only the list of the photo pairs (data and without its labels)

Here we provide our two best performing models, Alex Siamese (second-best) and VGG Siamese (best). They include caffemodel and prototxt files, described here.

Alex Siamese

Download alexsms.zip

VGG Siameses

Download vggsms.zip

For more information, please refer to the paper.
If you have any problems about the datasets and models, please contact .