# AMEBA82 人臉辨識使用scrfd模型，FPS只有2張

**URL:** <https://forum.amebaiot.com/t/ameba82-scrfd-fps-2/3965>\
**Category:** AI\
**Tags:** 中文, rtl8735b\
**Created:** [March 26, 2025, 1:32am UTC](https://forum.amebaiot.com/t/ameba82-scrfd-fps-2/3965 "2025-03-26T01:32:01Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![incifer](https://avatars.discourse-cdn.com/v4/letter/i/dc4da7/32.png) [@incifer](https://forum.amebaiot.com/u/incifer)\
**Post date:** [March 26, 2025, 1:32am UTC](https://forum.amebaiot.com/t/ameba82-scrfd-fps-2/3965/1 "2025-03-26T01:32:01Z")

</div>

我司拿 scrfd github([連結](https://github.com/deepinsight/insightface/tree/master/detection/scrfd)) 所提供的 預訓練模型(SCRFD\_500M\_KPS)轉換成nb，給 RTL8735 使用，FPS只有2張

如果使用範例: [RTSPFaceDetection.ino](https://github.com/Ameba-AIoT/ameba-arduino-pro2/tree/dev/Arduino_package/hardware/libraries/NeuralNetwork/examples/RTSPFaceDetection)，內建的人臉辨識FPS有7張，

請問我司是否有漏掉甚麼步驟導致 FPS 不足7張?

---

<div class="post-metadata">

**Author:** ![KevinKL](https://sea2.discourse-cdn.com/flex016/user_avatar/forum.amebaiot.com/kevinkl/32/2632_2.png) [@KevinKL](https://forum.amebaiot.com/u/KevinKL)\
**Post date:** [April 7, 2025, 8:51am UTC](https://forum.amebaiot.com/t/ameba82-scrfd-fps-2/3965/2 "2025-04-07T08:51:52Z")

</div>

您好

我們已把您的問題複製成功, 現在正嘗試尋找解決方案, 感謝您的耐心等待

---

<div class="post-metadata">

**Author:** ![KevinKL](https://sea2.discourse-cdn.com/flex016/user_avatar/forum.amebaiot.com/kevinkl/32/2632_2.png) [@KevinKL](https://forum.amebaiot.com/u/KevinKL)\
**Post date:** [April 7, 2025, 9:41am UTC](https://forum.amebaiot.com/t/ameba82-scrfd-fps-2/3965/3 "2025-04-07T09:41:51Z")

</div>

您好

```auto
// Lower resolution for NN processing
#define NNWIDTH 640
#define NNHEIGHT 640

```

麻煩您根據預訓練NN模型的resolution 調整 NNWIDTH 和 NNHEIGHT 參數, 重新燒入開發板再試看看

謝謝