基于Java深度学习库Deep Java Library 的一键抠图功能

九转成圣

发布于 2024-11-21 05:48:21

10500

代码可运行

文章被收录于专栏：csdncsdn

运行总次数：0

代码可运行

基于Java深度学习库Deep Java Library 的一键抠图功能

完整项目地址:https://download.csdn.net/download/weixin_41883161/89938027

一键抠图

一键抠图是一种图像处理技术，旨在自动将图像中的前景对象从背景中分离出来。它可以帮助用户快速、准确地实现抠图效果，无需手动绘制边界或进行复杂的图像编辑操作。一键抠图的原理通常基于计算机视觉和机器学习技术。它使用深度神经网络模型，通过训练大量的图像样本，学习如何识别和分离前景对象和背景。这些模型能够理解图像中的像素信息和上下文，并根据学习到的知识进行像素级别的分割。

动漫人物抠图

测试代码入口

public final class AnimeSegExample {

    private static final Logger logger = LoggerFactory.getLogger(AnimeSegExample.class);

    @SneakyThrows
    @Test
    public void test1() {
        Path imageFile = Paths.get("src/test/resources/anime-girl.jpg");
        Image img = BufferedImageFactory.getInstance().fromFile(imageFile);
        try (IsNetModel isNetModel = new IsNetModel("models/", "anime.onnx", 1, false, false, Device.cpu())) {
            Image segImg = isNetModel.predict(img);
            BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
            bufferedImage = ImageUtils.removeBg(bufferedImage);
            ImageUtils.saveBufferedImage(bufferedImage, "anime-girl.png", "src/test/output");
        }
    }
}

人物抠图

测试代码入口

public final class HumanSegExample {

    @SneakyThrows
    @Test
    public void test1() {
        Path imageFile = Paths.get("src/test/resources/fullbody2.jpg");
        Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);

        try (UNetHumanSegModel detector = new UNetHumanSegModel("models/", "human.onnx", 1, false, false, Device.cpu())) {
            Image segImg = detector.predict(img);
            BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
            bufferedImage = ImageUtils.removeBg(bufferedImage);
            ImageUtils.saveBufferedImage(bufferedImage, "fullbody2.jpg", "src/test/output/");
        }
    }
}

从人像上扣衣服

类位置top.aias.seg.ClothSegExample

扣上半身

代码入口

@SneakyThrows
@Test
public void test1() {
    Path imageFile = Paths.get("src/test/resources/full_body.jpg");
    Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);
    // clothCategory 4个值: 1,2,3,4  (1 上半身， 2 下半身, 3 连体衣, 4 所有）
    try (UNetClothSegModel detector = new UNetClothSegModel("models/", "cloth.onnx", 1, 1, false, Device.cpu())) {
        Image segImg = detector.predict(img);
        BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
        bufferedImage = ImageUtils.removeBg(bufferedImage);
        ImageUtils.saveBufferedImage(bufferedImage, "full_body_cloth_category_1.jpg", "src/test/output");
    }
}

扣下半身

代码入口

@SneakyThrows
@Test
public void test2() {
    Path imageFile = Paths.get("src/test/resources/full_body.jpg");
    Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);
    // clothCategory 4个值: 1,2,3,4  (1 上半身， 2 下半身, 3 连体衣, 4 所有）
    try (UNetClothSegModel detector = new UNetClothSegModel("models/", "cloth.onnx", 2, 1, false, Device.cpu())) {
        Image segImg = detector.predict(img);
        BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
        bufferedImage = ImageUtils.removeBg(bufferedImage);
        ImageUtils.saveBufferedImage(bufferedImage, "full_body_cloth_category_2.jpg", "src/test/output");
    }
}

扣全身

代码入口

@SneakyThrows
@Test
public void test4() {
    Path imageFile = Paths.get("src/test/resources/full_body.jpg");
    Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);
    // clothCategory 4个值: 1,2,3,4  (1 上半身， 2 下半身, 3 连体衣, 4 所有）
    try (UNetClothSegModel detector = new UNetClothSegModel("models/", "cloth.onnx", 4, 1, false, Device.cpu())) {
        Image segImg = detector.predict(img);
        BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
        bufferedImage = ImageUtils.removeBg(bufferedImage);
        ImageUtils.saveBufferedImage(bufferedImage, "full_body_cloth_category_4.jpg", "src/test/output");
    }
}

扣连衣裙

代码入口

@SneakyThrows
@Test
public void test3() {
    Path imageFile = Paths.get("src/test/resources/fullbody3.jpg");
    Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);
    // clothCategory 4个值: 1,2,3,4  (1 上半身， 2 下半身, 3 连体衣, 4 所有）
    try (UNetClothSegModel detector = new UNetClothSegModel("models/", "cloth.onnx", 3, 1, false, Device.cpu())) {
        Image segImg = detector.predict(img);
        BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
        bufferedImage = ImageUtils.removeBg(bufferedImage);
        ImageUtils.saveBufferedImage(bufferedImage, "fullbody3.jpg", "src/test/output");
    }
}

通用抠图

测试代码入口

@SneakyThrows
@Test
public void test1() {
    Path imageFile = Paths.get("src/test/resources/beauty.jpg");
    Image img = OpenCVImageFactory.getInstance().fromFile(imageFile);

    // u2net.onnx
    // u2netp.onnx
    // silueta.onnx
    try (UNetModel model = new UNetModel("models/", "u2net.onnx", 1, false, true, Device.cpu())) {
        Image segImg = model.predict(img);
        BufferedImage bufferedImage = OpenCVUtils.mat2Image((Mat) segImg.getWrappedImage());
        BufferedImage bImg = ImageUtils.removeBg(bufferedImage);
        ImageUtils.saveBufferedImage(bImg, "beauty.jpg", "src/test/output");
    }
}