# Pretrained models

**URL:** <https://community.drivendata.org/t/pretrained-models/7813>\
**Category:** Conser-vision Practice Area\
**Created:** [June 21, 2022, 9:51pm UTC](https://community.drivendata.org/t/pretrained-models/7813 "2022-06-21T21:51:45Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![sumatorikki](https://avatars.discourse-cdn.com/v4/letter/s/91b2a8/32.png) [@sumatorikki](https://community.drivendata.org/u/sumatorikki)\
**Post date:** [June 21, 2022, 9:51pm UTC](https://community.drivendata.org/t/pretrained-models/7813/1 "2022-06-21T21:51:45Z")

</div>

Hello, everyone,  
I am very pleased that you have organised such an interesting competition. I have a question, because in the regulations in 2 places was ambiguously specified.  
Is it allowed to use pretreated models e.g. ResNet50 and later tuning them?  
Does the model have to be 100% of your own invention?

---

<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:** [June 22, 2022, 3:22pm UTC](https://community.drivendata.org/t/pretrained-models/7813/2 "2022-06-22T15:22:58Z")

</div>

Hi @sumatorikki,

Yes, you can use pretrained models. The [benchmark tutorial](https://www.drivendata.co/blog/conservision-benchmark/) uses ResNet50, for example.

I hope this helps.