支持离线下载模型

2026-09-10 03:28:49 +00:00 · 2025-02-26 11:23:39 +08:00
parent e1aebbf37d
commit 685727e67c
15 changed files with 447 additions and 309 deletions
--- a/examples/face_offline.md
+++ b/examples/face_offline.md
@@ -0,0 +1,99 @@
+# SmartJavaAI离线下载模型案例
+
+**SmartJavaAI**如果未指定模型地址，系统将自动下载模型至本地。因此，无论模型是否通过离线方式下载，SmartJavaAI 最终都会在离线环境下运行模型。
+
+### 1. 安装人脸算法依赖
+
+在 Maven 项目的 `pom.xml` 中添加 SmartJavaAI的人脸算法依赖：
+
+```xml
+<dependencies>
+     <dependency>
+        <groupId>ink.numberone</groupId>
+        <artifactId>smartjavaai-face</artifactId>
+        <version>1.0.2</version>
+     </dependency>
+</dependencies>
+```
+
+### 2. 下载模型
+
+|         模型名称          |                           下载地址                           | 文件大小 |    适用场景    |
+| :-----------------------: | :----------------------------------------------------------: | :------: | :------------: |
+|        retinaface         | [下载](https://resources.djl.ai/test-models/pytorch/retinaface.zip) |  110MB   | 高精度人脸检测 |
+| ultralightfastgenericface | [下载](https://resources.djl.ai/test-models/pytorch/ultranet.zip) |  1.7MB   |  高速人脸检测  |
+|     featureExtraction     | [下载](https://resources.djl.ai/test-models/pytorch/face_feature.zip) |  104MB   |  人脸特征提取  |
+
+### 3. 人脸检测代码示例（离线下载模型）
+
+```java
+// 初始化配置
+ModelConfig config = new ModelConfig();
+config.setAlgorithmName("retinaface");//人脸算法模型，目前支持：retinaface及ultralightfastgenericface
+//config.setAlgorithmName("ultralightfastgenericface");//轻量模型
+config.setConfidenceThreshold(FaceConfig.DEFAULT_CONFIDENCE_THRESHOLD);//置信度阈值
+config.setMaxFaceCount(FaceConfig.MAX_FACE_LIMIT);//每张特征图保留的最大候选框数量
+//nms阈值:控制重叠框的合并程度,取值越低，合并越多重叠框（减少误检但可能漏检）；取值越高，保留更多框（增加检出但可能引入冗余）
+config.setNmsThresh(FaceConfig.NMS_THRESHOLD);
+//模型下载地址：
+//retinaface: https://resources.djl.ai/test-models/pytorch/retinaface.zip
+//ultralightfastgenericface: https://resources.djl.ai/test-models/pytorch/ultranet.zip
+//改为模型存放路径
+config.setModelPath("/Users/xxx/Documents/develop/face_model/retinaface.pt");
+//创建人脸算法
+FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceAlgorithm(config);
+//使用图片路径检测
+FaceDetectedResult result = currentAlgorithm.detect("src/main/resources/largest_selfie.jpg");
+logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
+//使用图片流检测
+File input = new File("src/main/resources/largest_selfie.jpg");
+//FaceDetectedResult result = currentAlgorithm.detect(new FileInputStream(input));
+//logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
+BufferedImage image = ImageIO.read(input);
+//创建保存路径
+Path imagePath = Paths.get("output").resolve("retinaface_detected.jpg");
+//绘制人脸框
+ImageUtils.drawBoundingBoxes(image, result, imagePath.toAbsolutePath().toString());
+```
+
+### 3. 人证核验示例（离线下载模型）
+
+人证核验步骤：
+
+（1）提取身份证人脸特征，
+
+（2）提取实时人脸特征
+
+（3）特征比对
+
+```java
+// 初始化配置
+ModelConfig config = new ModelConfig();
+config.setAlgorithmName("featureExtraction");
+//模型下载地址：https://resources.djl.ai/test-models/pytorch/face_feature.zip
+//改为模型存放路径
+config.setModelPath("/Users/xxx/Documents/develop/face_model/face_feature.pt");
+//创建脸算法
+FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceFeatureAlgorithm(config);
+//提取身份证人脸特征（图片仅供测试）
+float[] featureIdCard = currentAlgorithm.featureExtraction("src/main/resources/kana1.jpg");
+//提取身份证人脸特征（从图片流获取）
+//File input = new File("src/main/resources/kana1.jpg");
+//float[] featureIdCard = currentAlgorithm.featureExtraction(new FileInputStream(input));
+logger.info("身份证人脸特征：{}", JSONObject.toJSONString(featureIdCard));
+//提取实时人脸特征（图片仅供测试）
+float[] realTimeFeature = currentAlgorithm.featureExtraction("src/main/resources/kana2.jpg");
+logger.info("实时人脸特征：{}", JSONObject.toJSONString(realTimeFeature));
+if(realTimeFeature != null){
+    if(currentAlgorithm.calculSimilar(featureIdCard, realTimeFeature) > 0.8){
+        logger.info("人脸核验通过");
+    }else{
+        logger.info("人脸核验不通过");
+    }
+}
+```
+
+## 完整代码
+
+`📁 examples/src/main/java/smartai/examples/face`  
+ └── 📄[FaceDemo.java](https://github.com/geekwenjie/SmartJavaAI/blob/master/examples/src/main/java/smartai/examples/face/FaceDemo.java)  <sub>*（基于JDK11构建的完整可执行示例）*</sub>
--- a/examples/pom.xml
+++ b/examples/pom.xml
@@ -13,7 +13,7 @@
        <maven.compiler.target>11</maven.compiler.target>
        <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
        <smartjavaai.version>1.0.0-SNAPSHOT</smartjavaai.version>
-        <exec.mainClass>ai.djl.examples.inference.cv.ObjectDetection</exec.mainClass>
+        <exec.mainClass>smartai.examples.face.FaceDemo</exec.mainClass>
    </properties>


@@ -43,7 +43,7 @@
        <dependency>
            <groupId>ink.numberone</groupId>
            <artifactId>smartjavaai-face</artifactId>
-            <version>1.0.1</version>
+            <version>1.0.2</version>
        </dependency>

        <dependency>
--- a/examples/src/main/java/smartai/examples/face/FaceDemo.java
+++ b/examples/src/main/java/smartai/examples/face/FaceDemo.java
@@ -1,10 +1,7 @@
 package smartai.examples.face;

 import cn.smartjavaai.common.entity.Rectangle;
-import cn.smartjavaai.face.FaceAlgorithm;
-import cn.smartjavaai.face.FaceAlgorithmFactory;
-import cn.smartjavaai.face.FaceDetectedResult;
-import cn.smartjavaai.face.ModelConfig;
+import cn.smartjavaai.face.*;
 import com.alibaba.fastjson.JSONObject;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -26,14 +23,13 @@ import java.nio.file.Paths;
 */
 public class FaceDemo {

-    // 创建 Logger 实例
    private static final Logger logger = LoggerFactory.getLogger(FaceDemo.class);


    public static void main(String[] args) {
        try {
-            detectFace();
-            //verifyIDCard();
+            //detectFace();
+            verifyIDCard();
        } catch (Exception e) {
            e.printStackTrace();
        }
@@ -52,7 +48,7 @@ public class FaceDemo {
        FaceDetectedResult result = currentAlgorithm.detect("src/main/resources/largest_selfie.jpg");
        logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
        //使用图片流检测
-        //File input = new File("src/main/resources/largest_selfie.jpg");
+        File input = new File("src/main/resources/largest_selfie.jpg");
        //FaceDetectedResult result = currentAlgorithm.detect(new FileInputStream(input));
        //logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
        BufferedImage image = ImageIO.read(input);
@@ -92,7 +88,74 @@ public class FaceDemo {
     */
    public static void verifyIDCard() throws Exception {
        //创建脸算法
-        FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceAlgorithm();
+        FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceFeatureAlgorithm();
+        //提取身份证人脸特征（图片仅供测试）
+        float[] featureIdCard = currentAlgorithm.featureExtraction("src/main/resources/kana1.jpg");
+        //提取身份证人脸特征（从图片流获取）
+        //File input = new File("src/main/resources/kana1.jpg");
+        //float[] featureIdCard = currentAlgorithm.featureExtraction(new FileInputStream(input));
+        logger.info("身份证人脸特征：{}", JSONObject.toJSONString(featureIdCard));
+        //提取实时人脸特征（图片仅供测试）
+        float[] realTimeFeature = currentAlgorithm.featureExtraction("src/main/resources/kana2.jpg");
+        logger.info("实时人脸特征：{}", JSONObject.toJSONString(realTimeFeature));
+        if(realTimeFeature != null){
+            if(currentAlgorithm.calculSimilar(featureIdCard, realTimeFeature) > 0.8){
+                logger.info("人脸核验通过");
+            }else{
+                logger.info("人脸核验不通过");
+            }
+        }
+    }
+
+    /**
+     * 人脸检测（离线模型）
+     * 人脸模型：retinaface
+     * 特点：识别精度高，高速
+     * 应用场景：如监控摄像头、智能安防系统等需要高精度检测的场合
+     */
+    public static void detectFaceOffine() throws Exception {
+        // 初始化配置
+        ModelConfig config = new ModelConfig();
+        config.setAlgorithmName("retinaface");//人脸算法模型，目前支持：retinaface及ultralightfastgenericface
+        //config.setAlgorithmName("ultralightfastgenericface");//轻量模型
+        config.setConfidenceThreshold(FaceConfig.DEFAULT_CONFIDENCE_THRESHOLD);//置信度阈值
+        config.setMaxFaceCount(FaceConfig.MAX_FACE_LIMIT);//每张特征图保留的最大候选框数量
+        //nms阈值:控制重叠框的合并程度,取值越低，合并越多重叠框（减少误检但可能漏检）；取值越高，保留更多框（增加检出但可能引入冗余）
+        config.setNmsThresh(FaceConfig.NMS_THRESHOLD);
+        //模型下载地址：
+        //retinaface: https://resources.djl.ai/test-models/pytorch/retinaface.zip
+        //ultralightfastgenericface: https://resources.djl.ai/test-models/pytorch/ultranet.zip
+        //改为模型存放路径
+        config.setModelPath("/Users/xxx/Documents/develop/face_model/retinaface.pt");
+        //创建人脸算法
+        FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceAlgorithm(config);
+        //使用图片路径检测
+        FaceDetectedResult result = currentAlgorithm.detect("src/main/resources/largest_selfie.jpg");
+        logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
+        //使用图片流检测
+        File input = new File("src/main/resources/largest_selfie.jpg");
+        //FaceDetectedResult result = currentAlgorithm.detect(new FileInputStream(input));
+        //logger.info("人脸检测结果：{}", JSONObject.toJSONString(result));
+        BufferedImage image = ImageIO.read(input);
+        //创建保存路径
+        Path imagePath = Paths.get("output").resolve("retinaface_detected.jpg");
+        //绘制人脸框
+        ImageUtils.drawBoundingBoxes(image, result, imagePath.toAbsolutePath().toString());
+    }
+
+    /**
+     * 人证核验(离线模型)
+     * @throws Exception
+     */
+    public static void verifyIDCardOffine() throws Exception {
+        // 初始化配置
+        ModelConfig config = new ModelConfig();
+        config.setAlgorithmName("featureExtraction");
+        //模型下载地址：https://resources.djl.ai/test-models/pytorch/face_feature.zip
+        //改为模型存放路径
+        config.setModelPath("/Users/xxx/Documents/develop/face_model/face_feature.pt");
+        //创建脸算法
+        FaceAlgorithm currentAlgorithm = FaceAlgorithmFactory.createFaceFeatureAlgorithm(config);
        //提取身份证人脸特征（图片仅供测试）
        float[] featureIdCard = currentAlgorithm.featureExtraction("src/main/resources/kana1.jpg");
        //提取身份证人脸特征（从图片流获取）