Merge pull request #22 from geekwenjie/master

合并

README.md

@@ -242,6 +242,19 @@ SmartJavaAI是专为JAVA 开发者打造的一个功能丰富、开箱即用的
         </div>
       </td>
     </tr>
+ <tr>
+      <td>
+        <div align="left">
+          <p>零样本目标检测<br>(ZeroShot Object Detection)</p>
+      - YOLO-World 模型 <br>
+        </div>
+      </td>     
+      <td>
+        <div align="center">
+        <img src="https://cdn.jsdelivr.net/gh/geekwenjie/SmartJavaAI-Site/images/vision/yolo-world.png" height = "200px"/>
+        </div>
+      </td>
+    </tr>
   <tr>
       <td>
         <div align="left">
@@ -407,6 +420,8 @@ SmartJavaAI是专为JAVA 开发者打造的一个功能丰富、开箱即用的
   - 支持KINETICS400数据集中400个人类动作识别
 - **姿态估计**
   - 集成YOLOv8-pose、YOLOv11-pose等模型
+- **零样本目标检测**
+  - 集成YOLOv8s_worldv2、owlv2_base_patch16模型
 - **CLIP**
   - 支持提取图片及文本特征
   - 支持文搜图、图搜文、图搜图
@@ -483,7 +498,7 @@ SmartJavaAI是专为JAVA 开发者打造的一个功能丰富、开箱即用的
 <dependency>
     <groupId>cn.smartjavaai</groupId>
     <artifactId>all</artifactId>
-    <version>1.0.27</version>
+    <version>1.1.0</version>
 </dependency>
 ```
 
@@ -682,6 +697,14 @@ SmartJavaAI是专为JAVA 开发者打造的一个功能丰富、开箱即用的
 | YOLOV11-OBB | OnnxRuntime | Ultralytics在DOTAv1 数据集 上训练的模型、通过引入一个额外的角度来更准确地定位图像中的对象 | [Github](https://docs.ultralytics.com/zh/tasks/segment/) |
 ---
 
+#### 零样本目标检测模型
+
+| 模型名称        | 引擎      | 模型简介                           | 模型开源网站                                                   |
+|-------------|---------|--------------------------------|----------------------------------------------------------|
+| YOLOv8s-worldv2 | PyTorch | 可根据描述性文本检测图像中的任何物体 | [官网](https://docs.ultralytics.com/zh/models/yolo-world/) |
+| owlv2-base-patch16 | PyTorch | OWLv2是一种多模态模型，通过结合CLIP的骨干和ViT样的Transformer，实现零样本文本对象检测| [官网](https://huggingface.co/google/owlv2-base-patch16) |
+---
+
 #### 行人检测模型
 
 | 模型名称        | 引擎      | 模型开源网站     |

all/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <artifactId>all</artifactId>
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>${project.artifactId}</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>

bom/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <artifactId>bom</artifactId>
     <name>bom</name>
     <description>统一版本管理的 BOM 包，同时支持 import 和全量依赖</description>

common/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <name>common</name>

common/src/main/java/cn/smartjavaai/common/executor/GlobalExecutor.java

@@ -0,0 +1,53 @@
+package cn.smartjavaai.common.executor;
+
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.SynchronousQueue;
+import java.util.concurrent.ThreadPoolExecutor;
+import java.util.concurrent.TimeUnit;
+
+/**
+ * @author dwj
+ * @date 2025/11/26
+ */
+public class GlobalExecutor {
+
+    private static volatile ExecutorService executor;
+
+    public static ExecutorService getExecutor() {
+        if (executor == null) {
+            synchronized (GlobalExecutor.class) {
+                if (executor == null) {
+                    int cores = Runtime.getRuntime().availableProcessors();
+                    executor = new ThreadPoolExecutor(
+                            cores,
+                            cores * 2,
+                            60L, TimeUnit.SECONDS,
+                            new SynchronousQueue<>(),
+                            runnable -> {
+                                Thread t = new Thread(runnable);
+                                t.setDaemon(true); // 守护线程
+                                return t;
+                            },
+                            new ThreadPoolExecutor.DiscardOldestPolicy()
+                    );
+                }
+            }
+        }
+        return executor;
+    }
+
+    public static void shutdown() {
+        if (executor != null) {
+            executor.shutdown();
+            try {
+                if (!executor.awaitTermination(5, TimeUnit.SECONDS)) {
+                    executor.shutdownNow();
+                }
+            } catch (InterruptedException e) {
+                executor.shutdownNow();
+                Thread.currentThread().interrupt();
+            }
+        }
+    }
+
+}

examples/face-example/pom.xml

@@ -12,7 +12,7 @@
         <maven.compiler.source>11</maven.compiler.source>
         <maven.compiler.target>11</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <smartjavaai.version>1.0.27</smartjavaai.version>
+        <smartjavaai.version>1.1.0</smartjavaai.version>
         <!--如果打包运行，需要替换成你的main-->
         <exec.mainClass>smartai.examples.face.facedet.FaceDetDemo</exec.mainClass>
 
@@ -88,11 +88,10 @@
             <artifactId>face</artifactId>
         </dependency>
 
-
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-jni</artifactId>
-            <version>2.5.1-0.32.0</version>
+            <version>2.7.1-0.34.0</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -129,7 +128,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.windows-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -167,19 +166,45 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
+
+        <!-- linux aarch64 平台 (保留对应平台的配置，可以减小包大小)-->
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>javacpp</artifactId>
+            <version>${javacv.version}</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
         <dependency>
-            <groupId>ai.djl.pytorch</groupId>
-            <artifactId>pytorch-native-cpu-precxx11</artifactId>
-            <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
-            <scope>runtime</scope>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>ffmpeg</artifactId>
+            <version>6.1.1-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
         </dependency>
 
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>openblas</artifactId>
+            <version>0.3.26-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
 
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>opencv</artifactId>
+            <version>4.9.0-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
 
+        <dependency>
+            <groupId>ai.djl.pytorch</groupId>
+            <artifactId>pytorch-native-cpu</artifactId>
+            <classifier>${djl.platform.linux-aarch64}</classifier>
+            <version>2.7.1</version>
+            <scope>runtime</scope>
+        </dependency>
 
         <!-- macOS M系列 平台  (保留对应平台的配置，可以减小包大小)-->
         <dependency>
@@ -213,7 +238,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.osx-aarch64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 

examples/ocr-examples/pom.xml

@@ -12,7 +12,7 @@
         <maven.compiler.source>11</maven.compiler.source>
         <maven.compiler.target>11</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <smartjavaai.version>1.0.27</smartjavaai.version>
+        <smartjavaai.version>1.1.0</smartjavaai.version>
         <!--如果打包运行，需要替换成你的main-->
         <exec.mainClass>smartai.examples.ocr.common.OcrRecognizeDemo</exec.mainClass>
 
@@ -95,7 +95,7 @@
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-jni</artifactId>
-            <version>2.5.1-0.32.0</version>
+            <version>2.7.1-0.34.0</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -132,7 +132,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.windows-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -170,14 +170,43 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
+
+        <!-- linux aarch64 平台 (保留对应平台的配置，可以减小包大小)-->
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>javacpp</artifactId>
+            <version>${javacv.version}</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>ffmpeg</artifactId>
+            <version>6.1.1-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>openblas</artifactId>
+            <version>0.3.26-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>opencv</artifactId>
+            <version>4.9.0-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
-            <artifactId>pytorch-native-cpu-precxx11</artifactId>
-            <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <artifactId>pytorch-native-cpu</artifactId>
+            <classifier>${djl.platform.linux-aarch64}</classifier>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -214,7 +243,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.osx-aarch64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 

examples/speech-examples/pom.xml

@@ -12,7 +12,7 @@
         <maven.compiler.source>11</maven.compiler.source>
         <maven.compiler.target>11</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <smartjavaai.version>1.0.27</smartjavaai.version>
+        <smartjavaai.version>1.1.0</smartjavaai.version>
         <!--如果打包运行，需要替换成你的main-->
         <exec.mainClass>smartai.examples.speech.asr.common.OcrRecognizeDemo</exec.mainClass>
 

examples/translation-example/pom.xml

@@ -12,7 +12,7 @@
         <maven.compiler.source>11</maven.compiler.source>
         <maven.compiler.target>11</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <smartjavaai.version>1.0.27</smartjavaai.version>
+        <smartjavaai.version>1.1.0</smartjavaai.version>
         <!--如果打包运行，需要替换成你的main-->
         <exec.mainClass>smartai.examples.nlp.translation.TranslationDemo</exec.mainClass>
 
@@ -93,7 +93,7 @@
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-jni</artifactId>
-            <version>2.5.1-0.32.0</version>
+            <version>2.7.1-0.34.0</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -104,7 +104,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.windows-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -116,25 +116,25 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
+        <!-- linux aarch64 平台 (保留对应平台的配置，可以减小包大小)-->
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
-            <artifactId>pytorch-native-cpu-precxx11</artifactId>
-            <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <artifactId>pytorch-native-cpu</artifactId>
+            <classifier>${djl.platform.linux-aarch64}</classifier>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
-
         <!-- macOS M系列 平台  (保留对应平台的配置，可以减小包大小)-->
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.osx-aarch64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 

examples/translation-example/src/main/java/smartai/examples/nlp/translation/TranslationDemo.java

@@ -44,7 +44,7 @@ public class TranslationDemo {
         //指定翻译模型：NLLB,切换模型需同时修改modelEnum及modelPath
         config.setModelEnum(TranslationModeEnum.NLLB_MODEL);
         //指定模型路径，需将模型路径修改为本地的模型路径
-        config.setModelPath("/Users/xxx/Documents/develop/model/trans/traced_translation_cpu.pt");
+        config.setModelPath("/Users/wenjie/Documents/develop/model/translate/nllb/traced_translation_cpu.pt");
         config.setDevice(DeviceEnum.CPU);
         return TranslationModelFactory.getInstance().getModel(config);
     }

examples/vision-example/pom.xml

@@ -12,7 +12,7 @@
         <maven.compiler.source>11</maven.compiler.source>
         <maven.compiler.target>11</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <smartjavaai.version>1.0.27</smartjavaai.version>
+        <smartjavaai.version>1.1.0</smartjavaai.version>
         <!--如果打包运行，需要替换成你的main-->
         <exec.mainClass>smartai.examples.vision.ObjectDetectionDemo</exec.mainClass>
 
@@ -92,7 +92,7 @@
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-jni</artifactId>
-            <version>2.5.1-0.32.0</version>
+            <version>2.7.1-0.34.0</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -129,7 +129,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.windows-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -179,12 +179,11 @@
             <classifier>${javacv.platform.linux-x86_64}</classifier>
         </dependency>
 
-        <!--PyTorch离线平台依赖-->
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
@@ -203,14 +202,57 @@
             <version>1.9.1</version>
         </dependency>
 
+        <!-- linux aarch64 平台 (保留对应平台的配置，可以减小包大小)-->
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>javacpp</artifactId>
+            <version>${javacv.version}</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>ffmpeg</artifactId>
+            <version>6.1.1-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>openblas</artifactId>
+            <version>0.3.26-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
+        <dependency>
+            <groupId>org.bytedeco</groupId>
+            <artifactId>opencv</artifactId>
+            <version>4.9.0-1.5.10</version>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+        </dependency>
+
         <dependency>
             <groupId>ai.djl.pytorch</groupId>
-            <artifactId>pytorch-native-cpu-precxx11</artifactId>
-            <classifier>${djl.platform.linux-x86_64}</classifier>
-            <version>2.5.1</version>
+            <artifactId>pytorch-native-cpu</artifactId>
+            <classifier>${djl.platform.linux-aarch64}</classifier>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 
+        <dependency>
+            <groupId>ai.djl.tensorflow</groupId>
+            <artifactId>tensorflow-native-cpu</artifactId>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+            <scope>runtime</scope>
+            <version>2.16.1</version>
+        </dependency>
+        <dependency>
+            <groupId>ai.djl.mxnet</groupId>
+            <artifactId>mxnet-native-mkl</artifactId>
+            <classifier>${javacv.platform.linux-arm64}</classifier>
+            <scope>runtime</scope>
+            <version>1.9.1</version>
+        </dependency>
+
 
         <!-- macOS M系列 平台  (保留对应平台的配置，可以减小包大小)-->
         <dependency>
@@ -244,7 +286,7 @@
             <groupId>ai.djl.pytorch</groupId>
             <artifactId>pytorch-native-cpu</artifactId>
             <classifier>${djl.platform.osx-aarch64}</classifier>
-            <version>2.5.1</version>
+            <version>2.7.1</version>
             <scope>runtime</scope>
         </dependency>
 

examples/vision-example/src/main/java/smartai/examples/vision/ClsDemo.java

@@ -46,7 +46,7 @@ public class ClsDemo {
 
     public ClsModel getModel(){
         ClsModelConfig config = new ClsModelConfig();
-        //实例分割模型，切换模型需要同时修改modelEnum及modelPath
+        //切换模型需要同时修改modelEnum及modelPath
         config.setModelEnum(ClsModelEnum.YOLOV8);
         //模型所在路径，synset.txt也需要放在同目录下
         config.setModelPath("/Users/wenjie/Documents/develop/model/vision/cls/yolo11m-cls.onnx");

examples/vision-example/src/main/java/smartai/examples/vision/ZeroShotObjectDetectionDemo.java

@@ -0,0 +1,137 @@
+package smartai.examples.vision;
+
+import ai.djl.modality.cv.Image;
+
+import cn.smartjavaai.common.cv.SmartImageFactory;
+import cn.smartjavaai.common.entity.DetectionResponse;
+import cn.smartjavaai.common.entity.R;
+import cn.smartjavaai.common.enums.DeviceEnum;
+import cn.smartjavaai.common.utils.ImageUtils;
+import cn.smartjavaai.zeroshot.config.ZeroDetConfig;
+import cn.smartjavaai.zeroshot.enums.ZeroDetModelEnum;
+import cn.smartjavaai.zeroshot.model.ZeroDetModel;
+import cn.smartjavaai.zeroshot.model.ZeroDetModelFactory;
+import com.alibaba.fastjson.JSONObject;
+import lombok.extern.slf4j.Slf4j;
+import org.junit.BeforeClass;
+import org.junit.Test;
+
+import java.io.IOException;
+import java.io.OutputStream;
+import java.nio.file.Files;
+import java.nio.file.Path;
+import java.nio.file.Paths;
+
+/**
+ * 零样本目标检测
+ * @author dwj
+ */
+@Slf4j
+public class ZeroShotObjectDetectionDemo {
+
+
+    //设备类型
+    public static DeviceEnum device = DeviceEnum.CPU;
+
+    @BeforeClass
+    public static void beforeAll() throws IOException {
+        //修改缓存路径
+//        Config.setCachePath("/Users/xxx/smartjavaai_cache");
+    }
+
+    /**
+     * 获取零样本目标检测模型
+     */
+    public ZeroDetModel getModel(){
+        ZeroDetConfig config = new ZeroDetConfig();
+        //零样本目标检测模型，切换模型需要同时修改modelEnum及modelPath
+        config.setModelEnum(ZeroDetModelEnum.OWLV2_BASE_PATCH16);
+        //模型所在路径
+        config.setModelPath("/Users/wenjie/Documents/develop/model/vision/zero/owlv2-base-patch16");
+        config.setDevice(device);
+        //置信度阈值
+        config.setThreshold(0.5f);
+        return ZeroDetModelFactory.getInstance().getModel(config);
+    }
+
+
+
+    /**
+     * 零样本目标检测
+     * 特性：
+     * 1、零样本检测能力：无需针对特定类别进行训练，可直接通过文本查询检测新类别物体
+     * 2、开放词汇识别：能够识别训练时未见过的类别名称，突破传统检测模型的类别限制
+     * 3、多查询支持：支持同时使用多个文本查询进行目标检测，提高检测效率
+     */
+    @Test
+    public void zeroDetection(){
+        try {
+            ZeroDetModel detectorModel = getModel();
+            //创建Image对象，可以从文件、url、InputStream创建、BufferedImage、Base64创建，具体使用方法可以查看文档
+            Image image = SmartImageFactory.getInstance().fromFile(Paths.get("src/main/resources/zero/000000039769.jpg"));
+            //输入图片以及条件
+            R<DetectionResponse> result = detectorModel.detect(image, new String[]{"cat","remote control"});
+            if(result.isSuccess()){
+                log.info("零样本目标检测结果：{}", JSONObject.toJSONString(result.getData()));
+            }else{
+                log.info("零样本目标检测失败：{}", result.getMessage());
+            }
+        } catch (Exception e) {
+            e.printStackTrace();
+        }
+    }
+
+    /**
+     * 零样本目标检测并绘制检测结果
+     * 特性：
+     * 1、零样本检测能力：无需针对特定类别进行训练，可直接通过文本查询检测新类别物体
+     * 2、开放词汇识别：能够识别训练时未见过的类别名称，突破传统检测模型的类别限制
+     * 3、多查询支持：支持同时使用多个文本查询进行目标检测，提高检测效率
+     */
+    @Test
+    public void zeroDetectionAndDraw() {
+        try {
+            ZeroDetModel detectorModel = getModel();
+            String[] candidates = new String[]{"cat","remote control"};
+            //保存绘制后图片以及返回检测结果
+            R<DetectionResponse> result = detectorModel.detectAndDraw(candidates, "src/main/resources/zero/000000039769.jpg","output/cat_detected.png");
+            if(result.isSuccess()){
+                log.info("零样本目标检测结果：{}", JSONObject.toJSONString(result.getData()));
+            }else{
+                log.info("零样本目标检测失败：{}", result.getMessage());
+            }
+        } catch (Exception e) {
+            e.printStackTrace();
+        }
+
+    }
+
+    /**
+     * 零样本目标检测并绘制检测结果
+     * 特性：
+     * 1、零样本检测能力：无需针对特定类别进行训练，可直接通过文本查询检测新类别物体
+     * 2、开放词汇识别：能够识别训练时未见过的类别名称，突破传统检测模型的类别限制
+     * 3、多查询支持：支持同时使用多个文本查询进行目标检测，提高检测效率
+     */
+    @Test
+    public void zeroDetectionAndDraw2(){
+        try {
+            ZeroDetModel detectorModel = getModel();
+            //创建Image对象，可以从文件、url、InputStream创建、BufferedImage、Base64创建，具体使用方法可以查看文档
+            Image image = SmartImageFactory.getInstance().fromFile(Paths.get("src/main/resources/zero/000000039769.jpg"));
+            String[] candidates = new String[]{"cat","remote control"};
+            R<DetectionResponse> result = detectorModel.detectAndDraw(image, candidates);
+            if(result.isSuccess()){
+                log.info("零样本目标检测结果：{}", JSONObject.toJSONString(result.getData()));
+                //保存图片
+                ImageUtils.save(result.getData().getDrawnImage(), "output/cat_detected.png");
+            }else{
+                log.info("零样本目标检测失败：{}", result.getMessage());
+            }
+        } catch (Exception e) {
+            e.printStackTrace();
+        }
+
+    }
+
+}

face/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <artifactId>face</artifactId>
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>face</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>
@@ -26,7 +26,7 @@
 <!--        <maven.compiler.target>11</maven.compiler.target>-->
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <maven.test.skip>true</maven.test.skip>
-        <javacv.version>1.5.8</javacv.version>
+        <javacv.version>1.5.10</javacv.version>
         <javacv.ffmpeg.version>5.1.2-1.5.8</javacv.ffmpeg.version>
     </properties>
 
@@ -57,6 +57,8 @@
         </dependency>
 
 
+
+
     </dependencies>
 
     <build>

face/src/main/java/cn/smartjavaai/face/model/attribute/Seetaface6FaceAttributeModel.java

@@ -176,7 +176,7 @@ public class Seetaface6FaceAttributeModel implements FaceAttributeModel {
             imageData.data = BufferedImageUtils.getMatrixBGR(image);
             //检测人脸
             SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-            if(Objects.isNull(seetaResult)){
+            if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                 throw new FaceException("无人脸数据");
             }
             for(SeetaRect seetaRect : seetaResult){
@@ -456,7 +456,7 @@ public class Seetaface6FaceAttributeModel implements FaceAttributeModel {
             imageData.data = BufferedImageUtils.getMatrixBGR(image);
             //检测人脸
             SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-            if(Objects.isNull(seetaResult)){
+            if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                 throw new FaceException("无人脸数据");
             }
             SeetaPointF[] landmarks = new SeetaPointF[faceLandmarker.number()];
@@ -510,7 +510,7 @@ public class Seetaface6FaceAttributeModel implements FaceAttributeModel {
             imageData.data = ImageUtils.getMatrixBGR(image);
             //检测人脸
             SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-            if(Objects.isNull(seetaResult)){
+            if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                 throw new FaceException("无人脸数据");
             }
             for(SeetaRect seetaRect : seetaResult){
@@ -642,7 +642,7 @@ public class Seetaface6FaceAttributeModel implements FaceAttributeModel {
             imageData.data = ImageUtils.getMatrixBGR(image);
             //检测人脸
             SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-            if(Objects.isNull(seetaResult)){
+            if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                 throw new FaceException("无人脸数据");
             }
             SeetaPointF[] landmarks = new SeetaPointF[faceLandmarker.number()];

face/src/main/java/cn/smartjavaai/face/model/facedect/CommonFaceDetModel.java

@@ -9,6 +9,7 @@ import ai.djl.modality.cv.output.DetectedObjects;
 import ai.djl.repository.zoo.Criteria;
 import ai.djl.repository.zoo.ModelNotFoundException;
 import ai.djl.repository.zoo.ZooModel;
+import ai.djl.translate.TranslateException;
 import cn.smartjavaai.common.cv.SmartImageFactory;
 import cn.smartjavaai.common.entity.DetectionResponse;
 import cn.smartjavaai.common.entity.R;
@@ -50,6 +51,30 @@ public class CommonFaceDetModel implements FaceDetModel{
 
     private FaceDetConfig config;
 
+    @Override
+    public Predictor<Image, DetectedObjects> borrowPredictor() throws Exception {
+        if(predictorPool == null){
+            throw new FaceException("请先加载模型");
+        }
+        return predictorPool.borrowObject();
+    }
+
+    @Override
+    public void returnPredictor(Predictor<Image, DetectedObjects> predictor){
+        if (predictor != null) {
+            try {
+                predictorPool.returnObject(predictor); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+                try {
+                    predictor.close(); // 归还失败才销毁
+                } catch (Exception ex) {
+                    log.error("关闭Predictor失败", ex);
+                }
+            }
+        }
+    }
+
 
     /**
      * 加载模型

face/src/main/java/cn/smartjavaai/face/model/facedect/FaceDetModel.java

@@ -131,4 +131,20 @@ public interface FaceDetModel extends AutoCloseable{
     default void setFromFactory(boolean fromFactory){
         throw new UnsupportedOperationException("默认不支持该功能");
     }
+
+    /**
+     * 获取Predictor
+     * @return
+     */
+    default Predictor<Image, DetectedObjects> borrowPredictor() throws Exception{
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+    /**
+     * 归还Predictor
+     * @param predictor
+     */
+    default void returnPredictor(Predictor<Image, DetectedObjects> predictor){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
 }

face/src/main/java/cn/smartjavaai/face/model/facedect/FaceDetectManager.java

@@ -0,0 +1,99 @@
+package cn.smartjavaai.face.model.facedect;
+
+import ai.djl.inference.Predictor;
+import ai.djl.modality.cv.Image;
+import ai.djl.modality.cv.output.DetectedObjects;
+import cn.smartjavaai.common.entity.DetectionInfo;
+import cn.smartjavaai.common.entity.DetectionResponse;
+import cn.smartjavaai.common.entity.R;
+import cn.smartjavaai.face.exception.FaceException;
+import cn.smartjavaai.face.model.facedect.mtcnn.MtcnnPredictors;
+import cn.smartjavaai.face.seetaface.SeetaFace6FaceDetPredictors;
+import cn.smartjavaai.face.utils.FaceUtils;
+
+import java.util.Objects;
+
+/**
+ * @author dwj
+ * @date 2025/11/24
+ */
+public class FaceDetectManager implements AutoCloseable{
+
+
+    private FaceDetModel faceDetModel;
+
+    public FaceDetectManager(FaceDetModel faceDetModel) {
+        this.faceDetModel = faceDetModel;
+    }
+
+    private MtcnnPredictors mtcnnPredictors;
+
+    private SeetaFace6FaceDetPredictors seetaFace6FaceDetPredictors;
+
+    private Predictor<Image, DetectedObjects> commonPredictor;
+
+
+
+    public void borrowPredictors(){
+        try {
+            //mtcnn
+            if(faceDetModel instanceof MtcnnFaceDetModel){
+                MtcnnFaceDetModel mtcnnFaceDetModel = (MtcnnFaceDetModel) faceDetModel;
+                mtcnnPredictors = mtcnnFaceDetModel.borrowPredictors();
+            }else if(faceDetModel instanceof SeetaFace6FaceDetModel){
+                //SeetaFace6
+                SeetaFace6FaceDetModel seetaFace6FaceDetModel = (SeetaFace6FaceDetModel) faceDetModel;
+                seetaFace6FaceDetPredictors = seetaFace6FaceDetModel.borrowPredictors();
+            }else{
+                //其他通用模型
+                commonPredictor = faceDetModel.borrowPredictor();
+            }
+        } catch (Exception e) {
+            throw new FaceException("获取predictors异常", e);
+        }
+    }
+
+    public R<DetectionInfo> detectTopFace(Image image){
+        DetectionResponse detectionResponse = null;
+        try {
+            //mtcnn
+            if(faceDetModel instanceof MtcnnFaceDetModel){
+                MtcnnFaceDetModel mtcnnFaceDetModel = (MtcnnFaceDetModel) faceDetModel;
+                DetectedObjects detections = mtcnnFaceDetModel.detectCoreByPredictors(image, mtcnnPredictors);
+                detectionResponse = FaceUtils.convertToDetectionResponse(detections, image);
+            }else if(faceDetModel instanceof SeetaFace6FaceDetModel){
+                //SeetaFace6
+                SeetaFace6FaceDetModel seetaFace6FaceDetModel = (SeetaFace6FaceDetModel) faceDetModel;
+                detectionResponse = seetaFace6FaceDetModel.detectByPredictors(image, seetaFace6FaceDetPredictors);
+            }else{
+                DetectedObjects detections = commonPredictor.predict(image);
+                detectionResponse = FaceUtils.convertToDetectionResponse(detections, image);
+            }
+        } catch (Exception e) {
+            throw new FaceException("获取predictors异常", e);
+        }
+        if(Objects.isNull(detectionResponse) || Objects.isNull(detectionResponse.getDetectionInfoList()) || detectionResponse.getDetectionInfoList().isEmpty()){
+            return R.fail(R.Status.NO_FACE_DETECTED);
+        }
+        DetectionInfo detectionInfo = detectionResponse.getDetectionInfoList().get(0);
+        return R.ok(detectionInfo);
+    }
+
+
+    @Override
+    public void close(){
+        try {
+            //mtcnn
+            if(faceDetModel instanceof MtcnnFaceDetModel){
+                mtcnnPredictors.close();
+            }else if(faceDetModel instanceof SeetaFace6FaceDetModel){
+                //SeetaFace6
+                seetaFace6FaceDetPredictors.close();
+            }else{
+                faceDetModel.getPool().returnObject(commonPredictor);
+            }
+        } catch (Exception e) {
+            throw new FaceException("归还predictors异常", e);
+        }
+    }
+}

face/src/main/java/cn/smartjavaai/face/model/facedect/MtcnnFaceDetModel.java

@@ -221,52 +221,51 @@ public class MtcnnFaceDetModel extends CommonFaceDetModel{
     }
 
 
+    /**
+     * 使用MtcnnPredictors进行人脸检测
+     * @param image
+     * @param predictors
+     * @return
+     */
+    public DetectedObjects detectCoreByPredictors(Image image, MtcnnPredictors predictors){
+        Predictor<NDList, NDList> pNetPredictor = predictors.pNetPredictor;
+        Predictor<NDList, NDList> rNetPredictor = predictors.rNetPredictor;
+        Predictor<NDList, NDList> oNetPredictor = predictors.oNetPredictor;
+        try (NDManager manager = pNetModel.getNDManager().newSubManager();){
+            int h = image.getHeight();
+            int w = image.getWidth();
+            //第一阶段
+            NDList outputPnet = PNetModel.firstStage(manager, pNetPredictor, image);
 
-
-//    /**
-//     * 转换为FaceDetectedResult
-//     * @param mtcnnBatchResult
-//     * @return
-//     */
-//    public static DetectionResponse convertToDetectionResponse(MtcnnBatchResult mtcnnBatchResult){
-//        if(Objects.isNull(mtcnnBatchResult) || CollectionUtils.isEmpty(mtcnnBatchResult.boxes)
-//                || CollectionUtils.isEmpty(mtcnnBatchResult.points)
-//                || CollectionUtils.isEmpty(mtcnnBatchResult.probs)){
-//            return null;
-//        }
-//        DetectionResponse detectionResponse = new DetectionResponse();
-//        List<DetectionInfo> detectionInfoList = new ArrayList<DetectionInfo>();
-//
-//        NDArray boxes = mtcnnBatchResult.boxes.get(0);
-//        NDArray probs = mtcnnBatchResult.probs.get(0);
-//        NDArray points = mtcnnBatchResult.points.get(0);
-//
-//        if (DJLCommonUtils.isNDArrayEmpty(boxes) || DJLCommonUtils.isNDArrayEmpty(probs) || DJLCommonUtils.isNDArrayEmpty(points)){
-//            return null;
-//        }
-//        long numBoxes = boxes.getShape().get(0);
-//        for (int i = 0; i < numBoxes; i++) {
-//            float[] boxCoords = boxes.get(i).toFloatArray(); // [x1, y1, x2, y2]
-//            float score = probs.getFloat(i);
-//            NDArray pointND = points.get(i); // shape [5,2]
-//            float[] flatPoints = pointND.toFloatArray(); // 一维长度 10
-//            List<Point> keyPoints = new ArrayList<Point>();
-//            for (int p = 0; p < 5; p++) {
-//                keyPoints.add(new Point(flatPoints[p * 2], flatPoints[p * 2 + 1]));
-//            }
-//            int x = Math.round(boxCoords[0]);
-//            int y = Math.round(boxCoords[1]);
-//            int w = Math.round(boxCoords[2] - boxCoords[0]);
-//            int h = Math.round(boxCoords[3] - boxCoords[1]);
-//
-//            DetectionRectangle rectangle = new DetectionRectangle(x, y, w, h);
-//            FaceInfo faceInfo = new FaceInfo(keyPoints);
-//            DetectionInfo detectionInfo = new DetectionInfo(rectangle, score, faceInfo);
-//            detectionInfoList.add(detectionInfo);
-//        }
-//        detectionResponse.setDetectionInfoList(detectionInfoList);
-//        return detectionResponse;
-//    }
+            if(CollectionUtils.isEmpty(outputPnet)){
+                return DJLCommonUtils.buildEmptyDetectedObjects();
+            }
+            NDArray boxes = outputPnet.get(0);
+            NDArray image_inds = outputPnet.get(1);
+            NDArray imgs = outputPnet.get(2);
+            if(DJLCommonUtils.isNDArrayEmpty(boxes) || DJLCommonUtils.isNDArrayEmpty(image_inds) || DJLCommonUtils.isNDArrayEmpty(imgs)){
+                return DJLCommonUtils.buildEmptyDetectedObjects();
+            }
+            NDList pad = MtcnnUtils.pad(boxes, w, h);
+            //第二阶段
+            NDList outputRnet = RNetModel.secondStage(manager, rNetPredictor, imgs, boxes, pad, image_inds);
+            if(CollectionUtils.isEmpty(outputRnet)){
+                return DJLCommonUtils.buildEmptyDetectedObjects();
+            }
+            NDArray image_indsFiltered = outputRnet.get(0);
+            NDArray scoresFiltered = outputRnet.get(1);
+            boxes = outputRnet.get(2);
+            if(DJLCommonUtils.isNDArrayEmpty(boxes) || DJLCommonUtils.isNDArrayEmpty(image_indsFiltered) || DJLCommonUtils.isNDArrayEmpty(scoresFiltered)){
+                return DJLCommonUtils.buildEmptyDetectedObjects();
+            }
+            //第三阶段
+            MtcnnBatchResult oNetResult = ONetModel.thirdStage(manager, oNetPredictor, imgs, boxes, w, h, scoresFiltered, image_indsFiltered);
+            return FaceUtils.toDetectedObjects(oNetResult, w, h);
+        } catch (Exception e) {
+            e.printStackTrace();
+            throw new RuntimeException(e);
+        }
+    }
 
 
 
@@ -293,6 +292,57 @@ public class MtcnnFaceDetModel extends CommonFaceDetModel{
         return fromFactory;
     }
 
+
+    public MtcnnPredictors borrowPredictors() throws Exception {
+        if(pnetPredictorPool == null || rnetPredictorPool == null || onetPredictorPool == null){
+            return null;
+        }
+        Predictor<NDList, NDList> p = pnetPredictorPool.borrowObject();
+        Predictor<NDList, NDList> r = rnetPredictorPool.borrowObject();
+        Predictor<NDList, NDList> o = onetPredictorPool.borrowObject();
+        return new MtcnnPredictors(p, r, o, this);
+    }
+
+    public void returnPredictor(Predictor<NDList, NDList> pNetPredictor, Predictor<NDList, NDList> rNetPredictor, Predictor<NDList, NDList> oNetPredictor) {
+        if (pNetPredictor != null) {
+            try {
+                pnetPredictorPool.returnObject(pNetPredictor); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+                try {
+                    pNetPredictor.close(); // 归还失败才销毁
+                } catch (Exception ex) {
+                    log.error("关闭Predictor失败", ex);
+                }
+            }
+        }
+        if (rNetPredictor != null) {
+            try {
+                rnetPredictorPool.returnObject(rNetPredictor); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+                try {
+                    rNetPredictor.close(); // 归还失败才销毁
+                } catch (Exception ex) {
+                    log.error("关闭Predictor失败", ex);
+                }
+            }
+        }
+        if (oNetPredictor != null) {
+            try {
+                onetPredictorPool.returnObject(oNetPredictor); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+                try {
+                    oNetPredictor.close(); // 归还失败才销毁
+                } catch (Exception ex) {
+                    log.error("关闭Predictor失败", ex);
+                }
+            }
+        }
+    }
+
+
     @Override
     public void close() {
         if (fromFactory) {

face/src/main/java/cn/smartjavaai/face/model/facedect/SeetaFace6FaceDetModel.java

@@ -1,7 +1,9 @@
 package cn.smartjavaai.face.model.facedect;
 
 import ai.djl.engine.Engine;
+import ai.djl.inference.Predictor;
 import ai.djl.modality.cv.Image;
+import ai.djl.ndarray.NDList;
 import cn.smartjavaai.common.cv.SmartImageFactory;
 import cn.smartjavaai.common.entity.DetectionResponse;
 import cn.smartjavaai.common.entity.R;
@@ -13,7 +15,9 @@ import cn.smartjavaai.common.utils.ImageUtils;
 import cn.smartjavaai.face.config.FaceDetConfig;
 import cn.smartjavaai.face.exception.FaceException;
 import cn.smartjavaai.face.factory.FaceDetModelFactory;
+import cn.smartjavaai.face.model.facedect.mtcnn.MtcnnPredictors;
 import cn.smartjavaai.face.seetaface.NativeLoader;
+import cn.smartjavaai.face.seetaface.SeetaFace6FaceDetPredictors;
 import cn.smartjavaai.face.utils.FaceUtils;
 import com.seeta.pool.*;
 import com.seeta.sdk.*;
@@ -124,6 +128,27 @@ public class SeetaFace6FaceDetModel implements FaceDetModel{
         }
     }
 
+    public DetectionResponse detectByPredictors(Image image, SeetaFace6FaceDetPredictors predictors) {
+        SeetaImageData imageData = new SeetaImageData(image.getWidth(), image.getHeight(), 3);
+        imageData.data = ImageUtils.getMatrixBGR(image);
+        FaceDetector predictor = predictors.faceDetector;
+        FaceLandmarker faceLandmarker = predictors.faceLandmarker;
+        try {
+            SeetaRect[] seetaResult = predictor.Detect(imageData);
+            List<SeetaPointF[]> seetaPointFSList = new ArrayList<SeetaPointF[]>();
+            for(SeetaRect seetaRect : seetaResult){
+                //提取人脸的5点人脸标识
+                SeetaPointF[] pointFS = new SeetaPointF[faceLandmarker.number()];
+                faceLandmarker.mark(imageData, seetaRect, pointFS);
+                seetaPointFSList.add(pointFS);
+            }
+            return FaceUtils.convertToDetectionResponse(seetaResult, seetaPointFSList);
+        } catch (Exception e) {
+            throw new FaceException("目标检测错误", e);
+        }
+    }
+
+
     @Override
     public R<DetectionResponse> detectAndDraw(Image image) {
         R<DetectionResponse> result = detect(image);
@@ -276,6 +301,33 @@ public class SeetaFace6FaceDetModel implements FaceDetModel{
         return R.ok(drawnImage);
     }
 
+    public SeetaFace6FaceDetPredictors borrowPredictors() throws Exception {
+        if(faceDetectorPool == null || faceLandmarkerPool == null){
+            return null;
+        }
+        FaceDetector predictor = faceDetectorPool.borrowObject();
+        predictor.set(FaceDetector.Property.PROPERTY_THRESHOLD, config.getConfidenceThreshold() > 0 ? config.getConfidenceThreshold() : THRESHOLD);
+        FaceLandmarker faceLandmarker = faceLandmarkerPool.borrowObject();
+        return new SeetaFace6FaceDetPredictors(predictor, faceLandmarker, this);
+    }
+
+    public void returnPredictor(FaceDetector predictor, FaceLandmarker faceLandmarker) {
+        if (predictor != null) {
+            try {
+                faceDetectorPool.returnObject(predictor); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+            }
+        }
+        if (faceLandmarker != null) {
+            try {
+                faceLandmarkerPool.returnObject(faceLandmarker); //归还
+            } catch (Exception e) {
+                log.warn("归还Predictor失败", e);
+            }
+        }
+    }
+
 
 
 

face/src/main/java/cn/smartjavaai/face/model/facedect/mtcnn/MtcnnPredictors.java

@@ -0,0 +1,31 @@
+package cn.smartjavaai.face.model.facedect.mtcnn;
+
+import ai.djl.inference.Predictor;
+import ai.djl.ndarray.NDList;
+import cn.smartjavaai.face.model.facedect.MtcnnFaceDetModel;
+
+/**
+ * @author dwj
+ * @date 2025/11/24
+ */
+public class MtcnnPredictors implements AutoCloseable{
+
+    public Predictor<NDList, NDList> pNetPredictor;
+    public Predictor<NDList, NDList> rNetPredictor;
+    public Predictor<NDList, NDList> oNetPredictor;
+
+    // 标记是否由外部借用，用于控制 close 行为
+    private MtcnnFaceDetModel model;
+
+    public MtcnnPredictors(Predictor<NDList, NDList> p, Predictor<NDList, NDList> r, Predictor<NDList, NDList> o, MtcnnFaceDetModel m) {
+        this.pNetPredictor = p;
+        this.rNetPredictor = r;
+        this.oNetPredictor = o;
+        this.model = m;
+    }
+
+    @Override
+    public void close() throws Exception {
+        model.returnPredictor(pNetPredictor, rNetPredictor, oNetPredictor);
+    }
+}

face/src/main/java/cn/smartjavaai/face/model/liveness/CommonLivenessModel.java

@@ -6,6 +6,7 @@ import ai.djl.engine.Engine;
 import ai.djl.inference.Predictor;
 import ai.djl.modality.cv.Image;
 import ai.djl.modality.cv.ImageFactory;
+import ai.djl.modality.cv.output.DetectedObjects;
 import ai.djl.repository.zoo.Criteria;
 import ai.djl.repository.zoo.ModelNotFoundException;
 import ai.djl.repository.zoo.ZooModel;
@@ -28,8 +29,14 @@ import cn.smartjavaai.face.enums.LivenessModelEnum;
 import cn.smartjavaai.face.exception.FaceException;
 import cn.smartjavaai.face.factory.FaceDetModelFactory;
 import cn.smartjavaai.face.factory.LivenessModelFactory;
+import cn.smartjavaai.face.model.facedect.FaceDetectManager;
+import cn.smartjavaai.face.model.facedect.MtcnnFaceDetModel;
+import cn.smartjavaai.face.model.facedect.SeetaFace6FaceDetModel;
+import cn.smartjavaai.face.model.facedect.mtcnn.MtcnnPredictors;
 import cn.smartjavaai.face.model.liveness.criterial.LivenessCriteriaFactory;
 import cn.smartjavaai.face.model.liveness.translator.MiniVisionTranslator;
+import cn.smartjavaai.face.seetaface.SeetaFace6FaceDetPredictors;
+import cn.smartjavaai.face.utils.FaceUtils;
 import com.seeta.sdk.FaceAntiSpoofing;
 import lombok.extern.slf4j.Slf4j;
 import nu.pattern.OpenCV;
@@ -124,8 +131,12 @@ public class CommonLivenessModel implements LivenessDetModel{
         return detectVideo(new FFmpegFrameGrabber(videoPath));
     }
 
-    private R<LivenessResult> detectVideo(FFmpegFrameGrabber grabber) {
-        try {
+    protected R<LivenessResult> detectVideo(FFmpegFrameGrabber grabber) {
+        Predictor<Image, Float> predictor = null;
+        try (FaceDetectManager faceDetectManager = new FaceDetectManager(config.getDetectModel())){
+            //初始化predictors
+            faceDetectManager.borrowPredictors();
+            predictor = predictorPool.borrowObject();
             //滑动窗口
             Deque<Float> scoreWindow = new ArrayDeque<>();
             grabber.start();
@@ -147,7 +158,8 @@ public class CommonLivenessModel implements LivenessDetModel{
                         converterToMat = new OpenCVFrameConverter.ToOrgOpenCvCoreMat();
                     }
                     Mat mat = converterToMat.convert(frame);
-                    R<LivenessResult> livenessScore = detectTopFace(SmartImageFactory.getInstance().fromMat(mat));
+                    Image image = SmartImageFactory.getInstance().fromMat(mat);
+                    R<LivenessResult> livenessScore = detectVideoFrame(faceDetectManager, image, predictor);
                     mat.release();
                     if(!livenessScore.isSuccess()){
                         log.debug("第" + frameIndex + "帧处理失败：" + livenessScore.getMessage());
@@ -175,6 +187,24 @@ public class CommonLivenessModel implements LivenessDetModel{
             }
         } catch (Exception e) {
             throw new FaceException(e);
+        } finally {
+            if (predictor != null) {
+                try {
+                    predictorPool.returnObject(predictor); //归还
+                } catch (Exception e) {
+                    log.warn("归还Predictor失败", e);
+                    try {
+                        predictor.close(); // 归还失败才销毁
+                    } catch (Exception ex) {
+                        log.error("关闭Predictor失败", ex);
+                    }
+                }
+            }
+            try {
+                grabber.release();
+            } catch (FFmpegFrameGrabber.Exception e) {
+                throw new RuntimeException(e);
+            }
         }
         return R.fail(R.Status.Unknown);
     }
@@ -262,6 +292,40 @@ public class CommonLivenessModel implements LivenessDetModel{
         }
     }
 
+    private R<LivenessResult> detectVideoFrame(FaceDetectManager faceDetectManager, Image image, Predictor<Image, Float> predictor) {
+        //预处理图片
+        Image processedImage = null;
+        try {
+            //检测人脸
+            R<DetectionInfo> detectResult = faceDetectManager.detectTopFace(image);
+            if(!detectResult.isSuccess()){
+                return R.fail(detectResult.getCode(), detectResult.getMessage());
+            }
+            DetectionInfo detectionInfo = detectResult.getData();
+            if(config.getModelEnum() == LivenessModelEnum.IIC_FL_MODEL){
+                processedImage = new DJLImagePreprocessor(image, detectionInfo.getDetectionRectangle())
+                        .setExtendRatio(96f / 112f)
+                        .enableSquarePadding(true)
+                        .enableScaling(true)
+                        .setTargetSize(128)
+                        .enableCenterCrop(true)
+                        .setCenterCropSize(112)
+                        .process();
+            }
+            Float result = null;
+            if(processedImage != null){
+                result = predictor.predict(processedImage);
+                ImageUtils.releaseOpenCVMat(processedImage);
+            }else{
+                result = predictor.predict(image);
+            }
+            LivenessStatus status = result >= config.getRealityThreshold() ? LivenessStatus.LIVE : LivenessStatus.NON_LIVE;
+            return R.ok(new LivenessResult(status, result));
+        } catch (Exception e) {
+            throw new FaceException("活体检测错误", e);
+        }
+    }
+
     @Override
     public R<LivenessResult> detectTopFace(Image image) {
         R<DetectionResponse> faceDetectionResponse = config.getDetectModel().detect(image);

face/src/main/java/cn/smartjavaai/face/model/liveness/MiniVisionLivenessModel.java

@@ -10,6 +10,7 @@ import ai.djl.repository.zoo.Criteria;
 import ai.djl.repository.zoo.ModelNotFoundException;
 import ai.djl.repository.zoo.ZooModel;
 import ai.djl.training.util.ProgressBar;
+import cn.smartjavaai.common.cv.SmartImageFactory;
 import cn.smartjavaai.common.entity.*;
 import cn.smartjavaai.common.entity.face.FaceInfo;
 import cn.smartjavaai.common.entity.face.LivenessResult;
@@ -21,14 +22,19 @@ import cn.smartjavaai.common.preprocess.DJLImagePreprocessor;
 import cn.smartjavaai.common.utils.*;
 import cn.smartjavaai.face.config.LivenessConfig;
 import cn.smartjavaai.face.constant.MiniVisionConstant;
+import cn.smartjavaai.face.enums.LivenessModelEnum;
 import cn.smartjavaai.face.exception.FaceException;
 import cn.smartjavaai.face.factory.LivenessModelFactory;
+import cn.smartjavaai.face.model.facedect.FaceDetectManager;
 import cn.smartjavaai.face.model.liveness.translator.MiniVisionTranslator;
 import com.seeta.sdk.*;
 import lombok.extern.slf4j.Slf4j;
 import org.apache.commons.lang3.StringUtils;
 import org.apache.commons.pool2.ObjectPool;
 import org.apache.commons.pool2.impl.GenericObjectPool;
+import org.bytedeco.javacv.FFmpegFrameGrabber;
+import org.bytedeco.javacv.Frame;
+import org.bytedeco.javacv.OpenCVFrameConverter;
 import org.opencv.core.Mat;
 
 import javax.imageio.ImageIO;
@@ -59,6 +65,8 @@ public class MiniVisionLivenessModel extends CommonLivenessModel{
 
     private GenericObjectPool<Predictor<Image, float[]>> sePredictorPool;
 
+    private OpenCVFrameConverter.ToOrgOpenCvCoreMat converterToMat = null;
+
 
     /**
      * 模型策略
@@ -224,6 +232,147 @@ public class MiniVisionLivenessModel extends CommonLivenessModel{
         }
     }
 
+
+    protected R<LivenessResult> detectVideo(FFmpegFrameGrabber grabber) {
+        Predictor<Image, float[]> predictor = null;
+        Predictor<Image, float[]> sePredictor = null;
+        try (FaceDetectManager faceDetectManager = new FaceDetectManager(config.getDetectModel())){
+            //初始化predictors
+            faceDetectManager.borrowPredictors();
+            predictor = predictorPool.borrowObject();
+            sePredictor = sePredictorPool.borrowObject();
+            //滑动窗口
+            Deque<Float> scoreWindow = new ArrayDeque<>();
+            grabber.start();
+            // 获取视频总帧数
+            int totalFrames = grabber.getLengthInFrames();
+            log.debug("视频总帧数：{}，检测帧数：{}", totalFrames, config.getFrameCount());
+            if(totalFrames < config.getFrameCount()){
+                return R.fail(10001, "视频帧数低于检测帧数");
+            }
+            // 逐帧处理视频
+            for (int frameIndex = 0; frameIndex < totalFrames; frameIndex++) {
+                if(frameIndex >= config.getMaxVideoDetectFrames()){
+                    return R.fail(10002, "超出最大检测帧数：" + config.getMaxVideoDetectFrames());
+                }
+                // 获取当前帧
+                Frame frame = grabber.grabImage();
+                if (frame != null) {
+                    if(converterToMat == null){
+                        converterToMat = new OpenCVFrameConverter.ToOrgOpenCvCoreMat();
+                    }
+                    Mat mat = converterToMat.convert(frame);
+                    Image image = SmartImageFactory.getInstance().fromMat(mat);
+                    R<LivenessResult> livenessScore = detectVideoFrame(faceDetectManager, image, predictor, sePredictor);
+                    mat.release();
+                    if(!livenessScore.isSuccess()){
+                        log.debug("第" + frameIndex + "帧处理失败：" + livenessScore.getMessage());
+                        continue;
+                    }else{
+                        log.debug("第" + frameIndex + "帧活体检测结果：" + livenessScore);
+                        scoreWindow.add(livenessScore.getData().getScore());
+                    }
+                    // 如果累计检测帧数 >= 配置值，开始判断
+                    if (scoreWindow.size() >= config.getFrameCount()) {
+                        float avgScore = (float) scoreWindow.stream()
+                                .mapToDouble(Float::doubleValue)
+                                .average()
+                                .orElse(0.0);
+                        log.debug("滑动窗口平均得分: {}", avgScore);
+                        grabber.stop();
+                        LivenessStatus livenessStatus = avgScore > config.getRealityThreshold() ? LivenessStatus.LIVE : LivenessStatus.NON_LIVE;
+                        return R.ok(new LivenessResult(livenessStatus, avgScore));
+                    }
+                }
+            }
+            grabber.stop();
+            if(scoreWindow.size() < config.getFrameCount()){
+                return R.fail(1000, "有效帧数量不足，无法完成活体检测");
+            }
+        } catch (Exception e) {
+            throw new FaceException(e);
+        } finally {
+            if (predictor != null) {
+                try {
+                    predictorPool.returnObject(predictor); //归还
+                } catch (Exception e) {
+                    log.warn("归还Predictor失败", e);
+                    try {
+                        predictor.close(); // 归还失败才销毁
+                    } catch (Exception ex) {
+                        log.error("关闭Predictor失败", ex);
+                    }
+                }
+            }
+            if (sePredictor != null) {
+                try {
+                    sePredictorPool.returnObject(sePredictor); //归还
+                } catch (Exception e) {
+                    log.warn("归还Predictor失败", e);
+                    try {
+                        sePredictor.close(); // 归还失败才销毁
+                    } catch (Exception ex) {
+                        log.error("关闭Predictor失败", ex);
+                    }
+                }
+            }
+            try {
+                grabber.release();
+            } catch (FFmpegFrameGrabber.Exception e) {
+                throw new RuntimeException(e);
+            }
+        }
+        return R.fail(R.Status.Unknown);
+    }
+
+    private R<LivenessResult> detectVideoFrame(FaceDetectManager faceDetectManager, Image image, Predictor<Image, float[]> predictor, Predictor<Image, float[]> sePredictor) {
+        try {
+            //检测人脸
+            R<DetectionInfo> detectResult = faceDetectManager.detectTopFace(image);
+            if(!detectResult.isSuccess()){
+                return R.fail(detectResult.getCode(), detectResult.getMessage());
+            }
+            DetectionInfo detectionInfo = detectResult.getData();
+            float[] result = null;
+            float[] seResult = null;
+            //预处理图片
+            Image processedImage = new DJLImagePreprocessor(image, detectionInfo.getDetectionRectangle())
+                    .setExtendRatio(2.7f)
+                    .enableSquarePadding(true)
+                    .enableScaling(true)
+                    .setTargetSize(80)
+                    .process();
+            result = predictor.predict(processedImage);
+            ImageUtils.releaseOpenCVMat(processedImage);
+            //预处理图片
+            Image seProcessedImage = new DJLImagePreprocessor(image, detectionInfo.getDetectionRectangle())
+                    .setExtendRatio(4)
+                    .enableSquarePadding(true)
+                    .enableScaling(true)
+                    .setTargetSize(80)
+                    .process();
+            seResult = sePredictor.predict(seProcessedImage);
+            ImageUtils.releaseOpenCVMat(seProcessedImage);
+            if(Objects.isNull(result) && Objects.isNull(seResult)){
+                throw new FaceException("活体检测错误");
+            }
+            //计算结果
+            int maxIndex = ArrayUtils.sumAndFindMaxIndex(result, seResult, 3);
+            BigDecimal score = Objects.isNull(result) ? BigDecimal.ZERO : BigDecimal.valueOf(result[maxIndex]);
+            BigDecimal seScore = Objects.isNull(seResult) ? BigDecimal.ZERO : BigDecimal.valueOf(seResult[maxIndex]);
+            BigDecimal avgSocre = score.add(seScore).divide(BigDecimal.valueOf(2), 2, RoundingMode.HALF_UP);
+            //活体
+            if(maxIndex == 1){
+                LivenessStatus livenessStatus = avgSocre.floatValue() > config.getRealityThreshold() ? LivenessStatus.LIVE : LivenessStatus.NON_LIVE;
+                return R.ok(new LivenessResult(livenessStatus, avgSocre.floatValue()));
+            }else{//非活体
+                return R.ok(new LivenessResult(LivenessStatus.NON_LIVE, BigDecimal.ONE.subtract(avgSocre).floatValue()));
+            }
+        } catch (Exception e) {
+            throw new FaceException("活体检测错误", e);
+        }
+    }
+
     public GenericObjectPool<Predictor<Image, float[]>> getPredictorPool() {
         return predictorPool;
     }

face/src/main/java/cn/smartjavaai/face/model/liveness/Seetaface6LivenessModel.java

@@ -1,7 +1,9 @@
 package cn.smartjavaai.face.model.liveness;
 
 import ai.djl.engine.Engine;
+import ai.djl.inference.Predictor;
 import ai.djl.modality.cv.Image;
+import ai.djl.modality.cv.output.DetectedObjects;
 import cn.smartjavaai.common.cv.SmartImageFactory;
 import cn.smartjavaai.common.entity.*;
 import cn.smartjavaai.common.entity.face.FaceInfo;
@@ -15,6 +17,8 @@ import cn.smartjavaai.common.enums.face.LivenessStatus;
 import cn.smartjavaai.face.constant.LivenessConstant;
 import cn.smartjavaai.face.exception.FaceException;
 import cn.smartjavaai.face.factory.LivenessModelFactory;
+import cn.smartjavaai.face.model.facedect.FaceDetectManager;
+import cn.smartjavaai.face.model.facedect.SeetaFace6FaceDetModel;
 import cn.smartjavaai.face.seetaface.NativeLoader;
 import cn.smartjavaai.face.utils.FaceUtils;
 import cn.smartjavaai.face.utils.Seetaface6Utils;
@@ -23,6 +27,7 @@ import com.seeta.sdk.*;
 import lombok.extern.slf4j.Slf4j;
 import nu.pattern.OpenCV;
 import org.apache.commons.lang3.StringUtils;
+import org.apache.commons.pool2.impl.GenericObjectPool;
 import org.bytedeco.javacv.FFmpegFrameGrabber;
 import org.bytedeco.javacv.Frame;
 import org.bytedeco.javacv.Java2DFrameUtils;
@@ -60,6 +65,9 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
         if(StringUtils.isBlank(config.getModelPath())){
             throw new FaceException("modelPath is null");
         }
+        if(Objects.isNull(config.getDetectModel())){
+            throw new FaceException("未指定人脸检测模型");
+        }
         this.config = config;
         //加载依赖库
         NativeLoader.loadNativeLibraries(config.getDevice());
@@ -176,10 +184,41 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
         }
     }
 
+    private R<LivenessResult> detectVideoFrame(Image image, FaceDetectManager faceDetectManager, FaceAntiSpoofing faceAntiSpoofing) {
+        //检测人脸
+        R<DetectionInfo> detectResult = faceDetectManager.detectTopFace(image);
+        if(!detectResult.isSuccess()){
+            return R.fail(detectResult.getCode(), detectResult.getMessage());
+        }
+        DetectionInfo detectionInfo = detectResult.getData();
+        if(Objects.isNull(detectionInfo)){
+            return R.fail(R.Status.NO_FACE_DETECTED);
+        }
+        if(detectionInfo.getFaceInfo().getKeyPoints() == null || detectionInfo.getFaceInfo().getKeyPoints().isEmpty()){
+            return R.fail(1002,"人脸关键点keyPoints为空");
+        }
+        FaceAntiSpoofing.Status status = null;
+        try {
+            SeetaImageData imageData = new SeetaImageData(image.getWidth(), image.getHeight(), 3);
+            imageData.data = ImageUtils.getMatrixBGR(image);
+            SeetaRect seetaRect = Seetaface6Utils.convertToSeetaRect(detectionInfo.getDetectionRectangle());
+            SeetaPointF[] landmarks = Seetaface6Utils.convertToSeetaPointF(detectionInfo.getFaceInfo().getKeyPoints());
+            //检测视频
+            status = faceAntiSpoofing.PredictVideo(imageData, seetaRect, landmarks);
+            return R.ok(new LivenessResult(Seetaface6Utils.convertToLivenessStatus(status)));
+        } catch (Exception e) {
+            throw new FaceException("活体检测错误", e);
+        }
+    }
+
+
+
 
     private R<LivenessResult> detectVideo(FFmpegFrameGrabber grabber) {
         FaceAntiSpoofing faceAntiSpoofing = null;
-        try {
+        try (FaceDetectManager faceDetectManager = new FaceDetectManager(config.getDetectModel())){
+            //初始化predictors
+            faceDetectManager.borrowPredictors();
             faceAntiSpoofing = faceAntiSpoofingPool.borrowObject();
             //重置视频
             faceAntiSpoofing.ResetVideo();
@@ -194,14 +233,14 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
             // 逐帧处理视频
             for (int frameIndex = 0; frameIndex < totalFrames; frameIndex++) {
                 if(frameIndex >= config.getMaxVideoDetectFrames()){
-                    return R.fail(10002, "超出最大检测帧数：" + config.getMaxVideoDetectFrames());
+                    return R.fail(10002, "视频中未检测到人脸，超出最大检测帧数：" + config.getMaxVideoDetectFrames());
                 }
                 // 获取当前帧
                 Frame frame = grabber.grabImage();
                 if (frame != null) {
                     BufferedImage bufferedImage = Java2DFrameUtils.toBufferedImage(frame);
                     Image image = SmartImageFactory.getInstance().fromBufferedImage(bufferedImage);
-                    R<LivenessResult> livenessStatus = detectTopFace(image, false);
+                    R<LivenessResult> livenessStatus = detectVideoFrame(image, faceDetectManager, faceAntiSpoofing);
                     if(!livenessStatus.isSuccess()){
                         log.debug("第" + frameIndex + "帧处理失败：" + livenessStatus.getMessage());
                         continue;
@@ -225,10 +264,17 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
                     log.warn("归还Predictor失败", e);
                 }
             }
+            try {
+                grabber.release();
+            } catch (FFmpegFrameGrabber.Exception e) {
+                throw new RuntimeException(e);
+            }
         }
         return R.fail(1000, "有效帧数量不足，无法完成活体检测");
     }
 
+
+
     @Override
     public R<DetectionResponse> detect(Image image) {
         FaceAntiSpoofing faceAntiSpoofing = null;
@@ -246,7 +292,7 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
                 imageData.data = ImageUtils.getMatrixBGR(image);
                 //检测人脸
                 SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-                if(Objects.isNull(seetaResult)){
+                if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                     return R.fail(R.Status.NO_FACE_DETECTED);
                 }
                 for(SeetaRect seetaRect : seetaResult){
@@ -346,7 +392,7 @@ public class Seetaface6LivenessModel implements LivenessDetModel{
                 imageData.data = ImageUtils.getMatrixBGR(image);
                 //检测人脸
                 SeetaRect[] seetaResult = detectPredictor.Detect(imageData);
-                if(Objects.isNull(seetaResult)){
+                if(Objects.isNull(seetaResult) || seetaResult.length == 0){
                     return R.fail(R.Status.NO_FACE_DETECTED);
                 }
                 SeetaPointF[] landmarks = new SeetaPointF[faceLandmarker.number()];

face/src/main/java/cn/smartjavaai/face/seetaface/SeetaFace6FaceDetPredictors.java

@@ -0,0 +1,27 @@
+package cn.smartjavaai.face.seetaface;
+
+import cn.smartjavaai.face.model.facedect.SeetaFace6FaceDetModel;
+import com.seeta.sdk.FaceDetector;
+import com.seeta.sdk.FaceLandmarker;
+
+/**
+ * SeetaFace6 人脸检测Detector
+ * @author dwj
+ */
+public class SeetaFace6FaceDetPredictors implements AutoCloseable{
+
+    public FaceDetector faceDetector;
+    public FaceLandmarker faceLandmarker;
+    public SeetaFace6FaceDetModel model;
+
+    public SeetaFace6FaceDetPredictors(FaceDetector faceDetector, FaceLandmarker faceLandmarker, SeetaFace6FaceDetModel model) {
+        this.faceDetector = faceDetector;
+        this.faceLandmarker = faceLandmarker;
+        this.model = model;
+    }
+
+    @Override
+    public void close(){
+        model.returnPredictor(faceDetector, faceLandmarker);
+    }
+}

face/src/main/java/cn/smartjavaai/face/vector/core/MilvusClient.java

@@ -637,7 +637,8 @@ public class MilvusClient implements VectorDBClient {
 
             List<FaceVector> result = new ArrayList<>();
             for (QueryResultsWrapper.RowRecord row : records) {
-                String id = (String) row.get(VectorDBConstants.FieldNames.ID_FIELD);
+                Object idObj = row.get(VectorDBConstants.FieldNames.ID_FIELD);
+                String id = idObj != null ? idObj.toString() : null;
                 Object vectorObj = row.get(VectorDBConstants.FieldNames.VECTOR_FIELD);
                 float[] vector = null;
                 if (vectorObj instanceof List<?>) {

face/src/main/java/cn/smartjavaai/face/vector/core/SQLiteClient.java

@@ -2,6 +2,7 @@ package cn.smartjavaai.face.vector.core;
 
 import cn.hutool.core.util.IdUtil;
 import cn.smartjavaai.common.config.Config;
+import cn.smartjavaai.common.executor.GlobalExecutor;
 import cn.smartjavaai.common.utils.SimilarityUtil;
 import cn.smartjavaai.face.dao.FaceDao;
 import cn.smartjavaai.face.entity.FaceSearchParams;
@@ -23,12 +24,9 @@ import java.util.stream.Collectors;
 public class SQLiteClient implements VectorDBClient {
 
     private final FaceDao faceDao;
-    //private final List<FaceVector> memoryIndex = new CopyOnWriteArrayList<>();
     private final ConcurrentHashMap<String, FaceVector> memoryIndex = new ConcurrentHashMap<>();
     private int featureDimension; // 维度
 
-    private final ExecutorService executor = Executors.newFixedThreadPool(4);
-
     private SQLiteConfig config;
 
     /**
@@ -162,7 +160,7 @@ public class SQLiteClient implements VectorDBClient {
                     return similarity >= faceSearchParams.getThreshold() ?
                             new FaceSearchResult(vector.getId(), similarity, vector.getMetadata()) :
                             null;
-                }, executor))
+                }, GlobalExecutor.getExecutor()))
                 .collect(Collectors.toList());
 
         // 收集结果并过滤null
@@ -185,15 +183,7 @@ public class SQLiteClient implements VectorDBClient {
 
     @Override
     public void close() {
-        executor.shutdown();
-        try {
-            if (!executor.awaitTermination(5, TimeUnit.SECONDS)) {
-                executor.shutdownNow();
-            }
-        } catch (InterruptedException e) {
-            executor.shutdownNow();
-            Thread.currentThread().interrupt();
-        }
+
     }
 
     @Override

ocr/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <artifactId>ocr</artifactId>
@@ -42,7 +42,7 @@
         </dependency>
     </dependencies>
 
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>ocr</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>

pom.xml

@@ -7,7 +7,7 @@
     <name>SmartJavaAI</name>
     <groupId>cn.smartjavaai</groupId>
     <artifactId>smartjavaai-parent</artifactId>
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <packaging>pom</packaging>
     <description>SmartJavaAI</description>
     <modules>
@@ -26,7 +26,7 @@
         <maven.compiler.source>8</maven.compiler.source>
         <maven.compiler.target>8</maven.compiler.target>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-        <djl.version>0.32.0</djl.version>
+        <djl.version>0.34.0</djl.version>
     </properties>
 
     <dependencyManagement>

speech/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <artifactId>speech</artifactId>
@@ -57,7 +57,7 @@
     </dependencies>
 
 
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>speech</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>

translate/pom.xml

@@ -25,7 +25,7 @@
         </dependency>
     </dependencies>
 
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>translate</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>

vision/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>cn.smartjavaai</groupId>
         <artifactId>smartjavaai-parent</artifactId>
-        <version>1.0.27</version>
+        <version>1.1.0</version>
     </parent>
 
     <artifactId>vision</artifactId>
-    <version>1.0.27</version>
+    <version>1.1.0</version>
     <name>vision</name>
     <description>SmartJavaAI</description>
     <url>https://github.com/geekwenjie/SmartJavaAI</url>

vision/src/main/java/cn/smartjavaai/action/model/ActionRecModelFactory.java

@@ -55,7 +55,7 @@ public class ActionRecModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -64,7 +64,7 @@ public class ActionRecModelFactory {
      * @param config
      * @return
      */
-    private ActionRecModel createFaceDetModel(ActionRecModelConfig config) {
+    private ActionRecModel createModel(ActionRecModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/clip/model/ClipModel.java

@@ -66,8 +66,9 @@ public interface ClipModel extends AutoCloseable{
 
     /**
      * 图片特征比较
-     * @param image1 图1
-     * @param image2 图2
+     * @param image1
+     * @param image2
+     * @param scale
      * @return
      */
     default R<Float> compareImage(Image image1, Image image2, float scale){
@@ -115,10 +116,12 @@ public interface ClipModel extends AutoCloseable{
         throw new UnsupportedOperationException("默认不支持该功能");
     }
 
+
     /**
-     * 文本特征比较
-     * @param feature1 文本1
-     * @param feature2 文本2
+     * 特征比较
+     * @param feature1
+     * @param feature2
+     * @param scale
      * @return
      */
     default R<Float> compareFeatures(float[] feature1, float[] feature2, float scale){

vision/src/main/java/cn/smartjavaai/clip/model/ClipModelFactory.java

@@ -56,7 +56,7 @@ public class ClipModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -65,7 +65,7 @@ public class ClipModelFactory {
      * @param config
      * @return
      */
-    private ClipModel createFaceDetModel(ClipModelConfig config) {
+    private ClipModel createModel(ClipModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/cls/model/ClsModelFactory.java

@@ -55,7 +55,7 @@ public class ClsModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -64,7 +64,7 @@ public class ClsModelFactory {
      * @param config
      * @return
      */
-    private ClsModel createFaceDetModel(ClsModelConfig config) {
+    private ClsModel createModel(ClsModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/instanceseg/model/InstanceSegModelFactory.java

@@ -54,7 +54,7 @@ public class InstanceSegModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -63,7 +63,7 @@ public class InstanceSegModelFactory {
      * @param config
      * @return
      */
-    private InstanceSegModel createFaceDetModel(InstanceSegModelConfig config) {
+    private InstanceSegModel createModel(InstanceSegModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/obb/model/ObbDetModelFactory.java

@@ -54,7 +54,7 @@ public class ObbDetModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -63,7 +63,7 @@ public class ObbDetModelFactory {
      * @param config
      * @return
      */
-    private ObbDetModel createFaceDetModel(ObbDetModelConfig config) {
+    private ObbDetModel createModel(ObbDetModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/objectdetection/model/person/PersonDetModelFactory.java

@@ -56,7 +56,7 @@ public class PersonDetModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -65,7 +65,7 @@ public class PersonDetModelFactory {
      * @param config
      * @return
      */
-    private PersonDetModel createFaceDetModel(PersonDetModelConfig config) {
+    private PersonDetModel createModel(PersonDetModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/pose/model/PoseDetModelFactory.java

@@ -55,7 +55,7 @@ public class PoseDetModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -64,7 +64,7 @@ public class PoseDetModelFactory {
      * @param config
      * @return
      */
-    private PoseModel createFaceDetModel(PoseModelConfig config) {
+    private PoseModel createModel(PoseModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/semseg/model/SemSegModelFactory.java

@@ -54,7 +54,7 @@ public class SemSegModelFactory {
             throw new DetectionException("未配置模型");
         }
         return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
-            return createFaceDetModel(config);
+            return createModel(config);
         });
     }
 
@@ -63,7 +63,7 @@ public class SemSegModelFactory {
      * @param config
      * @return
      */
-    private SemSegModel createFaceDetModel(SemSegModelConfig config) {
+    private SemSegModel createModel(SemSegModelConfig config) {
         Class<?> clazz = registry.get(config.getModelEnum());
         if(clazz == null){
             throw new DetectionException("Unsupported model");

vision/src/main/java/cn/smartjavaai/zeroshot/config/ZeroDetConfig.java

@@ -0,0 +1,48 @@
+package cn.smartjavaai.zeroshot.config;
+
+import cn.smartjavaai.common.config.ModelConfig;
+import cn.smartjavaai.common.enums.DeviceEnum;
+import cn.smartjavaai.zeroshot.enums.ZeroDetModelEnum;
+import lombok.Data;
+
+import java.util.List;
+
+/**
+ * 零样本目标检测模型参数配置
+ *
+ * @author dwj
+ */
+@Data
+public class ZeroDetConfig extends ModelConfig {
+
+    /**
+     * 模型
+     */
+    private ZeroDetModelEnum modelEnum;
+
+
+    /**
+     * 模型路径
+     */
+    private String modelPath;
+
+    /**
+     * 置信度阈值
+     */
+    private float threshold = 0.3f;
+
+
+    public ZeroDetConfig() {
+    }
+
+    public ZeroDetConfig(ZeroDetModelEnum modelEnum, DeviceEnum device) {
+        this.modelEnum = modelEnum;
+        setDevice(device);
+    }
+
+    public ZeroDetConfig(ZeroDetModelEnum modelEnum) {
+        this.modelEnum = modelEnum;
+    }
+
+
+}

vision/src/main/java/cn/smartjavaai/zeroshot/criteria/ZeroDetCriteriaFactory.java

@@ -0,0 +1,53 @@
+package cn.smartjavaai.zeroshot.criteria;
+
+import ai.djl.Device;
+import ai.djl.huggingface.translator.ZeroShotObjectDetectionTranslatorFactory;
+import ai.djl.modality.cv.Image;
+import ai.djl.modality.cv.VisionLanguageInput;
+import ai.djl.modality.cv.output.DetectedObjects;
+import ai.djl.modality.cv.translator.YoloWorldTranslatorFactory;
+import ai.djl.repository.zoo.Criteria;
+import ai.djl.training.util.ProgressBar;
+import ai.djl.translate.TranslatorFactory;
+import cn.smartjavaai.common.enums.DeviceEnum;
+import cn.smartjavaai.zeroshot.config.ZeroDetConfig;
+import cn.smartjavaai.zeroshot.enums.ZeroDetModelEnum;
+import org.apache.commons.lang3.StringUtils;
+
+import java.nio.file.Paths;
+import java.util.Objects;
+import java.util.concurrent.ConcurrentHashMap;
+
+/**
+ * 零样本目标检测Criteria工厂
+ * @author dwj
+ */
+public class ZeroDetCriteriaFactory {
+
+
+    public static Criteria<VisionLanguageInput, DetectedObjects> createCriteria(ZeroDetConfig config) {
+        Device device = null;
+        if(!Objects.isNull(config.getDevice())){
+            device = config.getDevice() == DeviceEnum.CPU ? Device.cpu() : Device.gpu(config.getGpuId());
+        }
+        TranslatorFactory translatorFactory = null;
+        if(config.getModelEnum() == ZeroDetModelEnum.OWLV2_BASE_PATCH16){
+            translatorFactory = new ZeroShotObjectDetectionTranslatorFactory();
+        }else if(config.getModelEnum() == ZeroDetModelEnum.YOLOV8S_WORLDV2){
+            translatorFactory = new YoloWorldTranslatorFactory();
+        }
+        Criteria<VisionLanguageInput, DetectedObjects> criteria =
+                Criteria.builder()
+                        .setTypes(VisionLanguageInput.class, DetectedObjects.class)
+                        .optModelUrls(StringUtils.isNotBlank(config.getModelPath()) ? null :
+                                config.getModelEnum().getModelUri())
+                        .optModelPath(StringUtils.isNotBlank(config.getModelPath()) ? Paths.get(config.getModelPath()) : null)
+                        .optDevice(device)
+                        .optEngine(config.getModelEnum().getEngine())
+                        .optTranslatorFactory(translatorFactory)
+                        .optProgress(new ProgressBar())
+                        .build();
+
+        return criteria;
+    }
+}

vision/src/main/java/cn/smartjavaai/zeroshot/entity/DetectParams.java

@@ -0,0 +1,17 @@
+package cn.smartjavaai.zeroshot.entity;
+
+import lombok.Data;
+
+/**
+ * 检测参数
+ * @author dwj
+ */
+@Data
+public class DetectParams {
+
+    /**
+     * 置信度阈值
+     */
+    private float threshold = 0.3f;
+
+}

vision/src/main/java/cn/smartjavaai/zeroshot/enums/ZeroDetModelEnum.java

@@ -0,0 +1,45 @@
+package cn.smartjavaai.zeroshot.enums;
+
+/**
+ * 零样本目标检测模型枚举
+ * @author dwj
+ */
+public enum ZeroDetModelEnum {
+
+    YOLOV8S_WORLDV2("PyTorch", "djl://ai.djl.pytorch/yolov8s-worldv2"),
+    OWLV2_BASE_PATCH16("PyTorch", "djl://ai.djl.huggingface.pytorch/google/owlv2-base-patch16");
+
+
+    /**
+     * 根据名称获取枚举 (忽略大小写和下划线变体)
+     */
+    public static ZeroDetModelEnum fromName(String name) {
+        String formatted = name.trim().toUpperCase().replaceAll("[-_]", "");
+        for (ZeroDetModelEnum model : values()) {
+            if (model.name().replaceAll("_", "").equals(formatted)) {
+                return model;
+            }
+        }
+        throw new IllegalArgumentException("未知模型名称: " + name);
+    }
+
+    private final String modelUri;
+
+    /**
+     * 模型引擎
+     */
+    private final String engine;
+
+    ZeroDetModelEnum(String engine, String modelUri) {
+        this.modelUri = modelUri;
+        this.engine = engine;
+    }
+
+    public String getModelUri() {
+        return modelUri;
+    }
+
+    public String getEngine() {
+        return engine;
+    }
+}

vision/src/main/java/cn/smartjavaai/zeroshot/exception/ZeroDetException.java

@@ -0,0 +1,29 @@
+package cn.smartjavaai.zeroshot.exception;
+
+/**
+ * 零样本目标检测异常
+ * @author dwj
+ */
+public class ZeroDetException extends RuntimeException{
+
+    public ZeroDetException() {
+        super();
+    }
+
+    public ZeroDetException(String message, Throwable cause, boolean enableSuppression, boolean writableStackTrace) {
+        super(message, cause, enableSuppression, writableStackTrace);
+    }
+
+    public ZeroDetException(String message, Throwable cause) {
+        super(message, cause);
+    }
+
+    public ZeroDetException(String message) {
+        super(message);
+    }
+
+    public ZeroDetException(Throwable cause) {
+        super(cause);
+    }
+
+}

vision/src/main/java/cn/smartjavaai/zeroshot/model/CommonZeroDetModel.java

@@ -0,0 +1,162 @@
+package cn.smartjavaai.zeroshot.model;
+
+import ai.djl.MalformedModelException;
+import ai.djl.engine.Engine;
+import ai.djl.inference.Predictor;
+import ai.djl.modality.cv.Image;
+import ai.djl.modality.cv.VisionLanguageInput;
+import ai.djl.modality.cv.output.DetectedObjects;
+import ai.djl.repository.zoo.Criteria;
+import ai.djl.repository.zoo.ModelNotFoundException;
+import ai.djl.repository.zoo.ZooModel;
+import cn.smartjavaai.common.cv.SmartImageFactory;
+import cn.smartjavaai.common.entity.DetectionResponse;
+import cn.smartjavaai.common.entity.R;
+import cn.smartjavaai.common.pool.PredictorFactory;
+import cn.smartjavaai.objectdetection.exception.DetectionException;
+import cn.smartjavaai.vision.utils.DetectedObjectsFilter;
+import cn.smartjavaai.vision.utils.DetectorUtils;
+import cn.smartjavaai.zeroshot.config.ZeroDetConfig;
+import cn.smartjavaai.zeroshot.criteria.ZeroDetCriteriaFactory;
+import cn.smartjavaai.zeroshot.exception.ZeroDetException;
+import lombok.extern.slf4j.Slf4j;
+import org.apache.commons.pool2.impl.GenericObjectPool;
+
+import java.io.IOException;
+import java.nio.file.Files;
+import java.nio.file.Paths;
+import java.util.Objects;
+
+/**
+ * 零样本目标检测模型
+ * @author dwj
+ */
+@Slf4j
+public class CommonZeroDetModel implements ZeroDetModel {
+
+
+    private ZeroDetConfig config;
+
+    private ZooModel<VisionLanguageInput, DetectedObjects> model;
+
+    private GenericObjectPool<Predictor<VisionLanguageInput, DetectedObjects>> predictorPool;
+
+    @Override
+    public void loadModel(ZeroDetConfig config) {
+        if(Objects.isNull(config.getModelEnum())){
+            throw new DetectionException("未配置模型枚举");
+        }
+        Criteria<VisionLanguageInput, DetectedObjects> criteria = ZeroDetCriteriaFactory.createCriteria(config);
+        this.config = config;
+        try {
+            model = criteria.loadModel();
+            // 创建池子：每个线程独享 Predictor
+            this.predictorPool = new GenericObjectPool<>(new PredictorFactory<>(model));
+            int predictorPoolSize = config.getPredictorPoolSize();
+            if(config.getPredictorPoolSize() <= 0){
+                predictorPoolSize = Runtime.getRuntime().availableProcessors(); // 默认等于CPU核心数
+            }
+            predictorPool.setMaxTotal(predictorPoolSize);
+            log.debug("当前设备: " + model.getNDManager().getDevice());
+            log.debug("当前引擎: " + Engine.getInstance().getEngineName());
+            log.debug("模型推理器线程池最大数量: " + predictorPoolSize);
+        } catch (IOException | ModelNotFoundException | MalformedModelException e) {
+            throw new DetectionException("模型加载失败", e);
+        }
+    }
+
+    @Override
+    public R<DetectionResponse> detect(Image image, String[] candidates) {
+        DetectedObjects detectedObjects = detectCore(new VisionLanguageInput(image, candidates));
+        DetectionResponse detectionResponse = DetectorUtils.convertToDetectionResponse(detectedObjects, image);
+        return R.ok(detectionResponse);
+    }
+
+
+    /**
+     * 模型核心推理方法
+     * @param input
+     * @return
+     */
+    @Override
+    public DetectedObjects detectCore(VisionLanguageInput input) {
+        Predictor<VisionLanguageInput, DetectedObjects> predictor = null;
+        try {
+            predictor = predictorPool.borrowObject();
+            DetectedObjects detectedObjects = predictor.predict(input);
+            //过滤
+            if(Objects.nonNull(detectedObjects) && detectedObjects.getNumberOfObjects() > 0){
+                DetectedObjectsFilter detectedObjectsFilter = new DetectedObjectsFilter(null, config.getThreshold());
+                detectedObjects = detectedObjectsFilter.filter(detectedObjects);
+            }
+            return detectedObjects;
+        } catch (Exception e) {
+            throw new DetectionException("零样本目标检测错误", e);
+        }finally {
+            if (predictor != null) {
+                try {
+                    predictorPool.returnObject(predictor); //归还
+                    log.debug("释放资源");
+                } catch (Exception e) {
+                    log.warn("归还Predictor失败", e);
+                    try {
+                        predictor.close(); // 归还失败才销毁
+                    } catch (Exception ex) {
+                        log.error("关闭Predictor失败", ex);
+                    }
+                }
+            }
+        }
+    }
+
+    @Override
+    public R<DetectionResponse> detectAndDraw(Image image, String[] candidates) {
+        DetectedObjects detectedObjects = detectCore(new VisionLanguageInput(image, candidates));
+        image.drawBoundingBoxes(detectedObjects);
+        DetectionResponse detectionResponse = DetectorUtils.convertToDetectionResponse(detectedObjects, image);
+        detectionResponse.setDrawnImage(image);
+        return R.ok(detectionResponse);
+    }
+
+    @Override
+    public R<DetectionResponse> detectAndDraw(String[] candidates, String imagePath, String outputPath) {
+        try {
+            Image img = SmartImageFactory.getInstance().fromFile(Paths.get(imagePath));
+            DetectedObjects detectedObjects = detectCore(new VisionLanguageInput(img, candidates));
+            img.drawBoundingBoxes(detectedObjects);
+            img.save(Files.newOutputStream(Paths.get(outputPath)), "png");
+            DetectionResponse detectionResponse = DetectorUtils.convertToDetectionResponse(detectedObjects, img);
+            return R.ok(detectionResponse);
+        } catch (IOException e) {
+            throw new ZeroDetException(e);
+        }
+    }
+
+    private boolean fromFactory = false;
+
+    @Override
+    public void setFromFactory(boolean fromFactory) {
+        this.fromFactory = fromFactory;
+    }
+    public boolean isFromFactory() {
+        return fromFactory;
+    }
+
+    @Override
+    public void close() throws Exception {
+        try {
+            if (predictorPool != null) {
+                predictorPool.close();
+            }
+        } catch (Exception e) {
+            log.warn("关闭 predictorPool 失败", e);
+        }
+        try {
+            if (model != null) {
+                model.close();
+            }
+        } catch (Exception e) {
+            log.warn("关闭 model 失败", e);
+        }
+    }
+}

vision/src/main/java/cn/smartjavaai/zeroshot/model/ZeroDetModel.java

@@ -0,0 +1,50 @@
+package cn.smartjavaai.zeroshot.model;
+
+import ai.djl.modality.cv.Image;
+import ai.djl.modality.cv.VisionLanguageInput;
+import ai.djl.modality.cv.output.DetectedObjects;
+import cn.smartjavaai.common.entity.DetectionResponse;
+import cn.smartjavaai.common.entity.R;
+import cn.smartjavaai.zeroshot.config.ZeroDetConfig;
+
+/**
+ * 零样本目标检测模型
+ * @author dwj
+ */
+
+public interface ZeroDetModel extends AutoCloseable{
+
+
+    /**
+     * 加载模型
+     * @param config
+     */
+    void loadModel(ZeroDetConfig config);
+
+    /**
+     * 零样本目标检测
+     * @param image
+     * @return
+     */
+    default R<DetectionResponse> detect(Image image, String[] candidates){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+    default DetectedObjects detectCore(VisionLanguageInput input){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+    default R<DetectionResponse> detectAndDraw(Image image, String[] candidates){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+    default R<DetectionResponse> detectAndDraw(String[] candidates, String imagePath, String outputPath){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+    default void setFromFactory(boolean fromFactory){
+        throw new UnsupportedOperationException("默认不支持该功能");
+    }
+
+
+}

vision/src/main/java/cn/smartjavaai/zeroshot/model/ZeroDetModelFactory.java

@@ -0,0 +1,109 @@
+package cn.smartjavaai.zeroshot.model;
+
+import cn.smartjavaai.common.config.Config;
+import cn.smartjavaai.objectdetection.exception.DetectionException;
+import cn.smartjavaai.zeroshot.config.ZeroDetConfig;
+import cn.smartjavaai.zeroshot.enums.ZeroDetModelEnum;
+import lombok.extern.slf4j.Slf4j;
+
+import java.util.Map;
+import java.util.Objects;
+import java.util.concurrent.ConcurrentHashMap;
+
+/**
+ * 零样本目标检测 模型工厂
+ * @author dwj
+ */
+@Slf4j
+public class ZeroDetModelFactory {
+
+    // 使用 volatile 和双重检查锁定来确保线程安全的单例模式
+    private static volatile ZeroDetModelFactory instance;
+
+    private static final ConcurrentHashMap<ZeroDetModelEnum, ZeroDetModel> modelMap = new ConcurrentHashMap<>();
+
+    /**
+     * 模型注册表
+     */
+    private static final Map<ZeroDetModelEnum, Class<? extends ZeroDetModel>> registry =
+            new ConcurrentHashMap<>();
+
+
+    // 私有构造函数，防止外部创建实例
+    private ZeroDetModelFactory() {}
+
+    // 双重检查锁定的单例方法
+    public static ZeroDetModelFactory getInstance() {
+        if (instance == null) {
+            synchronized (ZeroDetModelFactory.class) {
+                if (instance == null) {
+                    instance = new ZeroDetModelFactory();
+                }
+            }
+        }
+        return instance;
+    }
+
+    /**
+     * 获取模型（通过配置）
+     * @param config
+     * @return
+     */
+    public ZeroDetModel getModel(ZeroDetConfig config) {
+        if(Objects.isNull(config) || Objects.isNull(config.getModelEnum())){
+            throw new DetectionException("未配置模型");
+        }
+        return modelMap.computeIfAbsent(config.getModelEnum(), k -> {
+            return createModel(config);
+        });
+    }
+
+    /**
+     * 使用ModelConfig创建模型
+     * @param config
+     * @return
+     */
+    private ZeroDetModel createModel(ZeroDetConfig config) {
+        Class<?> clazz = registry.get(config.getModelEnum());
+        if(clazz == null){
+            throw new DetectionException("Unsupported model");
+        }
+        ZeroDetModel model = null;
+        try {
+            model = (ZeroDetModel) clazz.newInstance();
+        } catch (InstantiationException | IllegalAccessException e) {
+            throw new DetectionException(e);
+        }
+        model.loadModel(config);
+        model.setFromFactory(true);
+        return model;
+    }
+
+
+    /**
+     * 注册模型
+     * @param modelEnum
+     * @param clazz
+     */
+    private static void registerAlgorithm(ZeroDetModelEnum modelEnum, Class<? extends ZeroDetModel> clazz) {
+        registry.put(modelEnum, clazz);
+    }
+
+
+    /**
+     * 移除缓存的模型
+     * @param modelEnum
+     */
+    public static void removeFromCache(ZeroDetModelEnum modelEnum) {
+        modelMap.remove(modelEnum);
+    }
+
+
+    // 初始化默认算法
+    static {
+        registerAlgorithm(ZeroDetModelEnum.YOLOV8S_WORLDV2, CommonZeroDetModel.class);
+        registerAlgorithm(ZeroDetModelEnum.OWLV2_BASE_PATCH16, CommonZeroDetModel.class);
+        log.debug("缓存目录：{}", Config.getCachePath());
+    }
+}
+