--- title: ImageNet 中的 Top-1 與 Top-5 accuracy date: 2020-03-10 is_modified: false disqus: cynthiahackmd categories: - "智慧計算 › 人工智慧" tags: - "ImageNet" - "AI/ML" - "煉丹常識" --- {%hackmd @CynthiaChuang/Github-Page-Theme %} <br> 在 ImageNet 的相關論文中很常見到兩種指標 ==Top-1== 與 ==Top-5==,記錄下所代表的涵義。 (我果然不擅長開頭阿:scream_cat:)。 <!--more--> <p class="illustration"> <img src="https://i.imgur.com/XeUtj5v.png" alt="模型概覽"> The Learning 模型概覽(圖片來源: <a href="https://keras.io/zh/applications/">Keras 中文文档</a>) </p> ## Top-1 accuracy 即對圖片進行預測時,機率最大的結果為正確答案,才視為正確。 ## Top-5 accuracy 對圖片進行預測時,機率排序後的前五的結果中包含正確答案,即視為正確。 ## 參考資料 1. Yannis Assael (2015-06-11)。[ImageNet: what is top-1 and top-5 error rate?](https://stats.stackexchange.com/questions/156471/imagenet-what-is-top-1-and-top-5-error-rate) 。檢自 Stack Exchange (2020-03-10)。 {%hackmd @CynthiaChuang/Github-Page-Footer %}
×
Sign in
Email
Password
Forgot password
or
By clicking below, you agree to our
terms of service
.
Sign in via Facebook
Sign in via Twitter
Sign in via GitHub
Sign in via Dropbox
Sign in with Wallet
Wallet (
)
Connect another wallet
New to HackMD?
Sign up