> 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/ko/versions/dataset-versions/image-augmentation/augmentation-types/flip-augmentation.md).

# 뒤집기 증강

이미지를 무작위로 세로 또는 가로로 뒤집기(반사)하여 모델이 피사체 방향에 덜 민감해지도록 합니다.

* **가로:** 이미지의 [NumPy 배열](https://numpy.org/) 좌우 방향으로.
* **세로:** 이미지의 [NumPy 배열](https://numpy.org/) 상하 방향으로.

## 예시

<figure><img src="/files/6498816ee7d5f94b7eed165a004bda392381db72" alt=""><figcaption><p>원본 이미지</p></figcaption></figure>

<figure><img src="/files/621a3ff3dc7a19637f5b07f4ff1ebe310a02e672" alt=""><figcaption><p>가로 뒤집기 증강</p></figcaption></figure>

<figure><img src="/files/82ec256a36f1b89c0396e7d2a6df6f3beb46996f" alt=""><figcaption><p>세로 뒤집기 증강</p></figcaption></figure>

## 자세히 알아보기

* [뒤집기 증강이 모델 성능을 어떻게 향상시키는지 확인해 보세요](https://blog.roboflow.com/how-flip-augmentation-improves-model-performance/)
