The image list used in "mergedNetRank.prototxt"

[ayantian]

Hi,
Thank you very much for sharing your code. One thing that puzzles me is that the code dose not provide the image list used in "mergedNetRank.prototxt". I wonder how you select pairs of images. I also read your paper, I noticed that you investigated two strategies for selecting pairs of images used in computing the ranking loss. One way is to select pairs of images with a relatively large difference in their average aesthetic scores. The other way is sample image pairs that have been scored by the same individual. I want to know how this relatively large difference is defined, whether there is a need to set a specific threshold for the difference in aesthetic scores, and how many pairs of images you used for train.
I am very looking forward to your help.
Best wishes!
Chen

[aimerykong]

Hi, Thanks for your interest. I didn't provide the list of image pairs because the txt file that stores the pairs is >200MB, meaning millions pairs are sampled for testing different strategies. As for sampling pairs with large margin, we chose those image pairs with at least 2 score difference, e.g. 3 vs. 5, 4 vs. 2, 3 vs. 1. Hope this helps. Regards, Shu

…

On Sun, Dec 10, 2017 at 9:44 PM, ayantian ***@***.***> wrote: Hi, Thank you very much for sharing your code. One thing that puzzles me is that the code dose not provide the image list used in "mergedNetRank.prototxt". I wonder how you select pairs of images. I also read your paper, I noticed that you investigated two strategies for selecting pairs of images used in computing the ranking loss. One way is to select pairs of images with a relatively large difference in their average aesthetic scores. The other way is sample image pairs that have been scored by the same individual. I want to know how this relatively large difference is defined, whether there is a need to set a specific threshold for the difference in aesthetic scores, and how many pairs of images you used for train. I am very looking forward to your help. Best wishes! Chen — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#15>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGKZJDh3BPMMvZjqx1eF0mx54d0jlpoMks5s_MEtgaJpZM4Q85D3> .