LPISP distance issues24

[jiaying96]

#24
HsinYingLee:

I think there are possible several settings can be used:

1. If you would like to compare with the ground truth set and other methods, say, a collection of N images. You can translate N target images from N source images. Then you can randomly sample M pairs out of this N images to calculate the diversity.
2. Other way to compare among different methods is, given N source images, for each image, you translate M of them and calculate a diversity score among these M target images. You then average the score of all N images.
We use the first setting in our experiments. Since there's no standard setting in this kind of experiment yet, I believe any setting that makes same should be okay.

I dont understand the way1 “randomly sample M pairs out of this N images to calculate the diversity”，in table2 "real images .448 ± .012", for each pair, is there a ground truth image A and a ground truth image B ? in table2 " DRIT .424 ± .010", for each pair, is there a translate image A ( use method DRIT and translate from ground truth image A )and a translate image B ( use method DRIT and translate from ground truth image B ?
@HsinYingLee

[KevinMarkVine]

#24
HsinYingLee:

I think there are possible several settings can be used:

1. If you would like to compare with the ground truth set and other methods, say, a collection of N images. You can translate N target images from N source images. Then you can randomly sample M pairs out of this N images to calculate the diversity.
2. Other way to compare among different methods is, given N source images, for each image, you translate M of them and calculate a diversity score among these M target images. You then average the score of all N images.
We use the first setting in our experiments. Since there's no standard setting in this kind of experiment yet, I believe any setting that makes same should be okay.

I dont understand the way1 “randomly sample M pairs out of this N images to calculate the diversity”，in table2 "real images .448 ± .012", for each pair, is there a ground truth image A and a ground truth image B ? in table2 " DRIT .424 ± .010", for each pair, is there a translate image A ( use method DRIT and translate from ground truth image A )and a translate image B ( use method DRIT and translate from ground truth image B ?
@HsinYingLee

I have the same problem, how did you solve it