> For the complete documentation index, see [llms.txt](https://docs.roboflow.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roboflow.com/datasets/ja/bjon/dataset-versions/image-augmentation/augmentation-types/cutout-augmentation.md).

# Cutout 拡張

画像の一部を黒いボックスでランダムにマスクし、物体の遮蔽に対してモデルを頑健にします。

{% hint style="info" %}
Cutout Augmentation は Enhanced Augmentation の一種で、 **プレミアム** 機能です。

プランとそれに関連する機能の最新情報については、 [料金ページ](https://roboflow.com/pricing).
{% endhint %}

Cutout は画像の一部を黒いボックスでランダムに置き換えます。Cutout を追加すると、オブジェクトの遮蔽に対してモデルの耐性を高めるのに役立ちます。

* 割合: 各 Cutout ボックスの画像に対するサイズの割合です。値が高いほど、より大きな遮蔽領域が生成されます。
* 数: 各画像に配置される Cutout ボックスの数です。値が高いほど、遮蔽の数が増えます。

{% hint style="danger" %}
多くの最新モデルは、トレーニング中に Cutout をオンライン拡張として適用します。2回適用すると望ましくない結果になる可能性があります。Roboflow 3.0 または YOLOv8 での使用は推奨しません。
{% endhint %}

## 例

<figure><img src="https://2416768182-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FNVfRxNsvCh4GOyqrjmzX%2Fuploads%2Fgit-blob-2347cf534e5a3427388b281e11477df72becafab%2Fimage%20(16).png?alt=media" alt=""><figcaption><p>元画像</p></figcaption></figure>

<figure><img src="https://2416768182-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FNVfRxNsvCh4GOyqrjmzX%2Fuploads%2Fgit-blob-b5cc23a9a9d3372e2800e7448a7c5e4476e4acd0%2Fimage%20(14).png?alt=media" alt=""><figcaption><p>Cutout を 10%、10 個に設定</p></figcaption></figure>
