About the holiday dataset.

[Niudong]

Hi, I'd like to raise a question. Here is the question. When I do some experiments on the Holiday dataset, I can not get the same results. There are almost 20 percents gaps between my results and yours. So I wonder if I can have one copy of your source code which contains the Holiday dataset part. Many thanks in advance.

[skamalas]

Hi Niudond,

This is weird - I assume you are able to reproduce the Oxford and Paris numbers? Unfortunately that code is not available. Just making sure, did you noticed this detail: As we say in the text, we used the rotated version of Holidays for your test, i.e. we manually turned all photos to have the same upright orientation. As I remember, though, such photos are not that many, and only this difference does not explain a 20% gap. It would be great if you also tried with rotated images and let us know.

Thanks, Yannis

[Niudong]

Hi.

You are right, I get almost the same numbers on the Oxford and Pairs datasets. I do not consider upright orientation the picture on the Holiday dataset.

Some details. For the limitation of GPU(11G), I set the input size with 768*768 because the Holiday image is too large, and I use the evaluation on the Holiday website, I get 63.372%. I wonder if you can give me some suggestions.

Many thanks in advance.

[pumpikano]

It seems there are two differences at least:

1. We corrected the orientation of all the images.
2. We did not resize the images when computing the features, if I remember correctly. I think for Holidays we computed the features on CPU, so avoided GPU memory issues.

But I think 20% is still surprising, even with these differences.

[Niudong]

Many thanks. I will do some experiments with your advice. By the way, I want to know what kind of evaluation method do you have on the Holiday dataset?

[skamalas]

As I remember we use the same evaluation protocol as in the other two datasets.

Another detail, whitening: As we say in the paper "We use the Oxford100k dataset for whitening on the Holidays."

[Niudong]

I will have another try. Thanks for your timely response.

[xiaozhi2015]

Are there any requirements for the PCA_whitening_data? Should they have the same distribution with the under test data? @skamalas Much thx!

[flyinsky323]

I will have another try. Thanks for your timely response.

Hello, mAP calculation on the Holiday dataset, have you implemented it? Can you tell me how to achieve it?

[ghost]

@Niudong Can you tell me how to achieve the method of map calculation on the Holiday dataset ?

Thanks in advance.