Palm Detection

2022 Spring Tongji Computer Vision Course Project

Method

1. Extract ROI —— Yolov5n (Run on Android Client)
2. Detect ROI —— MobileFaceNet (Run on Flask Server)

Paper: Towards Palmprint Verification On Smartphones (arxiv.org)

Flask Backend: LinzhouLi/PalmDetection-backend (github.com)

Deploy Yolov5

模型转换：

1. 训练好的pytorch模型导出为torchscript
2. 使用pnnx将torchscript格式转换为NCNN格式

图片预处理：

1. Resize (转换宽高至 $320\times 32k$ 或 $32k\times320$, $k\in \mathbb{Z}$)
2. Padding (以114填充添加的部分)
3. Normalize (除以255)

数据后处理：

Anchor & NMS

参考 nihui/ncnn-android-yolov5: The YOLOv5 object detection android example (github.com)

UI