# What is the use of Similarity normalization

**URL:** <https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669>\
**Category:** Image Similarity Challenge\
**Created:** [October 13, 2021, 3:44pm UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669 "2021-10-13T15:44:26Z")\
**Posts on this page:** 7\
**Page:** 1

<div class="post-metadata">

**Author:** ![ZaneRo](https://avatars.discourse-cdn.com/v4/letter/z/2bfe46/32.png) [@ZaneRo](https://community.drivendata.org/u/ZaneRo)\
**Post date:** [October 13, 2021, 3:44pm UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/1 "2021-10-13T15:44:26Z")

</div>

Hi,  
I don’t understand why use similarity normalization.  
If the Euclidean distance of the descriptor is used to measure the similarity, then the similarity is independent of other reference images.So we don’t need similarity normalization, right?

---

<div class="post-metadata">

**Author:** ![ZaneRo](https://avatars.discourse-cdn.com/v4/letter/z/2bfe46/32.png) [@ZaneRo](https://community.drivendata.org/u/ZaneRo)\
**Post date:** [October 15, 2021, 11:44am UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/2 "2021-10-15T11:44:50Z")

</div>

@mike-dd hi,can you help me?

---

<div class="post-metadata">

**Author:** ![mike-dd](https://avatars.discourse-cdn.com/v4/letter/m/f0a364/32.png) [@mike-dd](https://community.drivendata.org/u/mike-dd)\
**Post date:** [October 15, 2021, 1:43pm UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/3 "2021-10-15T13:43:05Z")

</div>

Hi @ZaneRo. Sure, can you tell me a bit more about the similarity normalization you’re referring to?

Is this something in the baseline or the competition paper? Or something we’ve said on the discussion forum? Please point me to a specific reference.

Thanks.

---

<div class="post-metadata">

**Author:** ![ZaneRo](https://avatars.discourse-cdn.com/v4/letter/z/2bfe46/32.png) [@ZaneRo](https://community.drivendata.org/u/ZaneRo)\
**Post date:** [October 16, 2021, 6:49am UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/4 "2021-10-16T06:49:33Z")

</div>

It’s in the competition paper: “The 2021 Image Similarity Dataset and Challenge”

> **[The 2021 Image Similarity Dataset and Challenge](https://arxiv.org/abs/2106.09672)**
>
> This paper introduces a new benchmark for large-scale image similarity
> detection. This benchmark is used for the Image Similarity Challenge at
> NeurIPS'21 (ISC2021). The goal is to determine whether a query image is a
> modified copy of any image in a...

---

<div class="post-metadata">

**Author:** ![ZaneRo](https://avatars.discourse-cdn.com/v4/letter/z/2bfe46/32.png) [@ZaneRo](https://community.drivendata.org/u/ZaneRo)\
**Post date:** [October 18, 2021, 1:15am UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/5 "2021-10-18T01:15:46Z")

</div>

@mike-dd hi, above is my reply.

---

<div class="post-metadata">

**Author:** ![mike-dd](https://avatars.discourse-cdn.com/v4/letter/m/f0a364/32.png) [@mike-dd](https://community.drivendata.org/u/mike-dd)\
**Post date:** [October 18, 2021, 4:49pm UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/6 "2021-10-18T16:49:24Z")

</div>

Got it, thanks for clarifying, @ZaneRo.

There is no requirement to use similarity normalization. The baseline models described in the competition paper use similarity normalization to improve the accuracy of the image pair scores, but you don’t have to do the same.

It sounds like there might be some confusion about why similarity normalization is being used. Similarity normalization is not intended to ensure independence from other reference images. But in applying similarity normalization, the baseline models must ensure that they are still following competition rules, and they do this by computing the normalization factor using the training set.

I hope this helps.

---

<div class="post-metadata">

**Author:** ![ZaneRo](https://avatars.discourse-cdn.com/v4/letter/z/2bfe46/32.png) [@ZaneRo](https://community.drivendata.org/u/ZaneRo)\
**Post date:** [October 19, 2021, 1:37am UTC](https://community.drivendata.org/t/what-is-the-use-of-similarity-normalization/6669/7 "2021-10-19T01:37:53Z")

</div>

ok,it helps. Thank you!