[Backend] Add OCR、Seg、 KeypointDetection、Matting、 ernie-3.0 and adaface models for XPU Deploy (#960)

* [FlyCV] Bump up FlyCV -> official release 1.0.0

* add seg models for XPU

* add ocr model for XPU

* add matting

* add matting python

* fix infer.cc

* add keypointdetection support for XPU

* Add adaface support for XPU

* add ernie-3.0

* fix doc

Co-authored-by: DefTruth <qiustudent_r@163.com>
Co-authored-by: DefTruth <31974251+DefTruth@users.noreply.github.com>

examples/text/ernie-3.0/cpp/README.md

@@ -35,6 +35,8 @@ tar xvfz ernie-3.0-medium-zh-afqmc.tgz
 # GPU Inference
 ./seq_cls_infer_demo --device gpu --model_dir ernie-3.0-medium-zh-afqmc
 
+# XPU 推理
+./seq_cls_infer_demo --device xpu --model_dir ernie-3.0-medium-zh-afqmc
 ```
 The result returned after running is as follows：
 ```bash

examples/text/ernie-3.0/cpp/seq_cls_infer.cc

@@ -32,7 +32,7 @@ const char sep = '/';
 DEFINE_string(model_dir, "", "Directory of the inference model.");
 DEFINE_string(vocab_path, "", "Path of the vocab file.");
 DEFINE_string(device, "cpu",
-              "Type of inference device, support 'cpu' or 'gpu'.");
+              "Type of inference device, support 'cpu', 'xpu' or 'gpu'.");
 DEFINE_string(backend, "onnx_runtime",
               "The inference runtime backend, support: ['onnx_runtime', "
               "'paddle', 'openvino', 'tensorrt', 'paddle_tensorrt']");
@@ -55,7 +55,16 @@ void PrintUsage() {
 }
 
 bool CreateRuntimeOption(fastdeploy::RuntimeOption* option) {
-  if (FLAGS_device == "gpu") {
+  std::string model_path = FLAGS_model_dir + sep + "infer.pdmodel";
+  std::string param_path = FLAGS_model_dir + sep + "infer.pdiparams";
+  fastdeploy::FDINFO << "model_path = " << model_path
+                     << ", param_path = " << param_path << std::endl;
+  option->SetModelPath(model_path, param_path);
+
+  if (FLAGS_device == "xpu") {
+    option->UseXpu();
+    return true;
+  } else if (FLAGS_device == "gpu") {
     option->UseGpu();
   } else if (FLAGS_device == "cpu") {
     option->UseCpu();
@@ -97,11 +106,7 @@ bool CreateRuntimeOption(fastdeploy::RuntimeOption* option) {
                         << FLAGS_backend << "'" << std::endl;
     return false;
   }
-  std::string model_path = FLAGS_model_dir + sep + "infer.pdmodel";
-  std::string param_path = FLAGS_model_dir + sep + "infer.pdiparams";
-  fastdeploy::FDINFO << "model_path = " << model_path
-                     << ", param_path = " << param_path << std::endl;
-  option->SetModelPath(model_path, param_path);
+  
   return true;
 }
 

examples/text/ernie-3.0/python/README.md

@@ -22,7 +22,7 @@ pip install -r requirements.txt
 
 ### A Quick Start
 
-The following example shows how to employ FastDeploy library to complete Python predictive deployment of ERNIE 3.0 Medium model on [AFQMC Dataset](https://bj.bcebos.com/paddlenlp/datasets/afqmc_public.zip)of CLUE Benchmark for text classification tasks. 
+The following example shows how to employ FastDeploy library to complete Python predictive deployment of ERNIE 3.0 Medium model on [AFQMC Dataset](https://bj.bcebos.com/paddlenlp/datasets/afqmc_public.zip)of CLUE Benchmark for text classification tasks.
 
 ```bash
 
@@ -40,6 +40,9 @@ python seq_cls_infer.py --device cpu --model_dir ernie-3.0-medium-zh-afqmc
 # GPU Inference
 python seq_cls_infer.py --device gpu --model_dir ernie-3.0-medium-zh-afqmc
 
+# XPU Inference
+python seq_cls_infer.py --device xpu --model_dir ernie-3.0-medium-zh-afqmc
+
 ```
 The result returned after running is as follows:
 

examples/text/ernie-3.0/python/seq_cls_infer.py

@@ -35,8 +35,8 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        choices=['gpu', 'cpu'],
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        choices=['gpu', 'cpu', 'xpu'],
+        help="Type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--backend",
         type=str,
@@ -94,6 +94,10 @@ class ErnieForSequenceClassificationPredictor(object):
         model_path = os.path.join(args.model_dir, "infer.pdmodel")
         params_path = os.path.join(args.model_dir, "infer.pdiparams")
         option.set_model_path(model_path, params_path)
+        if args.device == 'xpu':
+            option.use_xpu()
+            option.use_paddle_lite_backend()
+            return fd.Runtime(option)
         if args.device == 'cpu':
             option.use_cpu()
         else:

examples/vision/faceid/adaface/cpp/README.md

@@ -51,6 +51,13 @@ tar zxvf mobilefacenet_adaface.tgz -C ./
               test_lite_focal_arcface_2.JPG \
               2
 
+# XPU推理
+./infer_demo mobilefacenet_adaface/mobilefacenet_adaface.pdmodel \
+              mobilefacenet_adaface/mobilefacenet_adaface.pdiparams \
+              test_lite_focal_arcface_0.JPG \
+              test_lite_focal_arcface_1.JPG \
+              test_lite_focal_arcface_2.JPG \
+              3
 ```
 
 运行完成可视化结果如下图所示

examples/vision/faceid/adaface/cpp/infer.cc

@@ -47,6 +47,43 @@ void CpuInfer(const std::string &model_file, const std::string &params_file,
             << ", Cosine 02:" << cosine02 << std::endl;
 }
 
+void XpuInfer(const std::string &model_file, const std::string &params_file,
+              const std::vector<std::string> &image_file) {
+  auto option = fastdeploy::RuntimeOption();
+  option.UseXpu();
+  auto model = fastdeploy::vision::faceid::AdaFace(model_file, params_file);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  cv::Mat face0 = cv::imread(image_file[0]);
+  cv::Mat face1 = cv::imread(image_file[1]);
+  cv::Mat face2 = cv::imread(image_file[2]);
+
+  fastdeploy::vision::FaceRecognitionResult res0;
+  fastdeploy::vision::FaceRecognitionResult res1;
+  fastdeploy::vision::FaceRecognitionResult res2;
+
+  if ((!model.Predict(&face0, &res0)) || (!model.Predict(&face1, &res1)) ||
+      (!model.Predict(&face2, &res2))) {
+    std::cerr << "Prediction Failed." << std::endl;
+  }
+
+  std::cout << "Prediction Done!" << std::endl;
+
+  std::cout << "--- [Face 0]:" << res0.Str();
+  std::cout << "--- [Face 1]:" << res1.Str();
+  std::cout << "--- [Face 2]:" << res2.Str();
+
+  float cosine01 = fastdeploy::vision::utils::CosineSimilarity(
+      res0.embedding, res1.embedding, model.l2_normalize);
+  float cosine02 = fastdeploy::vision::utils::CosineSimilarity(
+      res0.embedding, res2.embedding, model.l2_normalize);
+  std::cout << "Detect Done! Cosine 01: " << cosine01
+            << ", Cosine 02:" << cosine02 << std::endl;
+}
+
 void GpuInfer(const std::string &model_file, const std::string &params_file,
               const std::vector<std::string> &image_file) {
   auto option = fastdeploy::RuntimeOption();
@@ -134,7 +171,7 @@ int main(int argc, char *argv[]) {
                  "test_lite_focal_AdaFace_2.JPG 0"
               << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with xpu."
               << std::endl;
     return -1;
   }
@@ -147,6 +184,8 @@ int main(int argc, char *argv[]) {
     GpuInfer(argv[1], argv[2], image_files);
   } else if (std::atoi(argv[6]) == 2) {
     TrtInfer(argv[1], argv[2], image_files);
+  } else if (std::atoi(argv[6]) == 3) {
+    CpuInfer(argv[1], argv[2], image_files);
   }
   return 0;
 }

examples/vision/faceid/adaface/python/README.md

@@ -45,6 +45,14 @@ python infer.py --model mobilefacenet_adaface/mobilefacenet_adaface.pdmodel \
                 --face_negative test_lite_focal_arcface_2.JPG \
                  --device gpu \
                  --use_trt True
+
+# XPU推理
+python infer.py --model mobilefacenet_adaface/mobilefacenet_adaface.pdmodel \
+                --params_file mobilefacenet_adaface/mobilefacenet_adaface.pdiparams \
+                --face test_lite_focal_arcface_0.JPG \
+                --face_positive test_lite_focal_arcface_1.JPG \
+                --face_negative test_lite_focal_arcface_2.JPG \
+                 --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/faceid/adaface/python/infer.py

@@ -39,7 +39,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        help="Type of inference device, support 'cpu' , 'xpu' or 'gpu'.")
     parser.add_argument(
         "--use_trt",
         type=ast.literal_eval,
@@ -54,6 +54,9 @@ def build_option(args):
     if args.device.lower() == "gpu":
         option.use_gpu()
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+
     if args.use_trt:
         option.use_trt_backend()
         option.set_trt_input_shape("data", [1, 3, 112, 112])

examples/vision/keypointdetection/det_keypoint_unite/cpp/README.md

@@ -33,6 +33,8 @@ wget https://bj.bcebos.com/paddlehub/fastdeploy/000000018491.jpg
 ./infer_demo PP_PicoDet_V2_S_Pedestrian_320x320_infer PP_TinyPose_256x192_infer 000000018491.jpg 1
 # GPU上TensorRT推理
 ./infer_demo PP_PicoDet_V2_S_Pedestrian_320x320_infer PP_TinyPose_256x192_infer 000000018491.jpg 2
+# XPU推理
+./infer_demo PP_PicoDet_V2_S_Pedestrian_320x320_infer PP_TinyPose_256x192_infer 000000018491.jpg 3
 ```
 
 运行完成可视化结果如下图所示

examples/vision/keypointdetection/det_keypoint_unite/cpp/det_keypoint_unite_infer.cc

@@ -66,6 +66,55 @@ void CpuInfer(const std::string& det_model_dir,
             << std::endl;
 }
 
+void XpuInfer(const std::string& det_model_dir,
+              const std::string& tinypose_model_dir,
+              const std::string& image_file) {
+  auto option = fastdeploy::RuntimeOption();
+  option.UseXpu();
+  auto det_model_file = det_model_dir + sep + "model.pdmodel";
+  auto det_params_file = det_model_dir + sep + "model.pdiparams";
+  auto det_config_file = det_model_dir + sep + "infer_cfg.yml";
+  auto det_model = fastdeploy::vision::detection::PicoDet(
+      det_model_file, det_params_file, det_config_file, option);
+  if (!det_model.Initialized()) {
+    std::cerr << "Detection Model Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto tinypose_model_file = tinypose_model_dir + sep + "model.pdmodel";
+  auto tinypose_params_file = tinypose_model_dir + sep + "model.pdiparams";
+  auto tinypose_config_file = tinypose_model_dir + sep + "infer_cfg.yml";
+  auto tinypose_model = fastdeploy::vision::keypointdetection::PPTinyPose(
+      tinypose_model_file, tinypose_params_file, tinypose_config_file, option);
+  if (!tinypose_model.Initialized()) {
+    std::cerr << "TinyPose Model Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  fastdeploy::vision::KeyPointDetectionResult res;
+
+  auto pipeline =
+      fastdeploy::pipeline::PPTinyPose(
+          &det_model, &tinypose_model);
+  pipeline.detection_model_score_threshold = 0.5;
+  if (!pipeline.Predict(&im, &res)) {
+    std::cerr << "TinyPose Prediction Failed." << std::endl;
+    return;
+  } else {
+    std::cout << "TinyPose Prediction Done!" << std::endl;
+  }
+  // 输出预测框结果
+  std::cout << res.Str() << std::endl;
+
+  // 可视化预测结果
+  auto vis_im =
+      fastdeploy::vision::VisKeypointDetection(im, res, 0.2);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "TinyPose visualized result saved in ./vis_result.jpg"
+            << std::endl;
+}
+
 void GpuInfer(const std::string& det_model_dir,
               const std::string& tinypose_model_dir,
               const std::string& image_file) {
@@ -180,7 +229,7 @@ int main(int argc, char* argv[]) {
                  "./test.jpeg 0"
               << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with xpu."
               << std::endl;
     return -1;
   }
@@ -191,6 +240,8 @@ int main(int argc, char* argv[]) {
     GpuInfer(argv[1], argv[2], argv[3]);
   } else if (std::atoi(argv[4]) == 2) {
     TrtInfer(argv[1], argv[2], argv[3]);
+  } else if (std::atoi(argv[4]) == 3) {
+    XpuInfer(argv[1], argv[2], argv[3]);
   }
   return 0;
 }

examples/vision/keypointdetection/det_keypoint_unite/python/README.md

@@ -25,6 +25,8 @@ python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infe
 python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device gpu
 # GPU上使用TensorRT推理 （注意：TensorRT推理第一次运行，有序列化模型的操作，有一定耗时，需要耐心等待）
 python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device gpu --use_trt True
+# XPU推理
+python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/keypointdetection/det_keypoint_unite/python/det_keypoint_unite_infer.py

@@ -19,7 +19,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="type of inference device, support 'cpu' or 'gpu'.")
+        help="type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--use_trt",
         type=ast.literal_eval,
@@ -47,6 +47,9 @@ def build_tinypose_option(args):
     if args.device.lower() == "gpu":
         option.use_gpu()
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+
     if args.use_trt:
         option.use_trt_backend()
         option.set_trt_input_shape("image", [1, 3, 256, 192])

examples/vision/keypointdetection/tiny_pose/cpp/README.md

@@ -32,6 +32,8 @@ wget https://bj.bcebos.com/paddlehub/fastdeploy/hrnet_demo.jpg
 ./infer_tinypose_demo PP_TinyPose_256x192_infer hrnet_demo.jpg 1
 # GPU上TensorRT推理
 ./infer_tinypose_demo PP_TinyPose_256x192_infer hrnet_demo.jpg 2
+# XPU推理
+./infer_tinypose_demo PP_TinyPose_256x192_infer hrnet_demo.jpg 3
 ```
 
 运行完成可视化结果如下图所示

examples/vision/keypointdetection/tiny_pose/cpp/pptinypose_infer.cc

@@ -53,6 +53,40 @@ void CpuInfer(const std::string& tinypose_model_dir,
             << std::endl;
 }
 
+void XpuInfer(const std::string& tinypose_model_dir,
+              const std::string& image_file) {
+  auto tinypose_model_file = tinypose_model_dir + sep + "model.pdmodel";
+  auto tinypose_params_file = tinypose_model_dir + sep + "model.pdiparams";
+  auto tinypose_config_file = tinypose_model_dir + sep + "infer_cfg.yml";
+  auto option = fastdeploy::RuntimeOption();
+  option.UseXpu();
+  auto tinypose_model = fastdeploy::vision::keypointdetection::PPTinyPose(
+      tinypose_model_file, tinypose_params_file, tinypose_config_file, option);
+  if (!tinypose_model.Initialized()) {
+    std::cerr << "TinyPose Model Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  fastdeploy::vision::KeyPointDetectionResult res;
+  if (!tinypose_model.Predict(&im, &res)) {
+    std::cerr << "TinyPose Prediction Failed." << std::endl;
+    return;
+  } else {
+    std::cout << "TinyPose Prediction Done!" << std::endl;
+  }
+  // 输出预测框结果
+  std::cout << res.Str() << std::endl;
+
+  // 可视化预测结果
+  auto tinypose_vis_im =
+      fastdeploy::vision::VisKeypointDetection(im, res, 0.5);
+  cv::imwrite("tinypose_vis_result.jpg", tinypose_vis_im);
+  std::cout << "TinyPose visualized result saved in ./tinypose_vis_result.jpg"
+            << std::endl;
+}
+
+
 void GpuInfer(const std::string& tinypose_model_dir,
               const std::string& image_file) {
   auto option = fastdeploy::RuntimeOption();
@@ -129,7 +163,7 @@ int main(int argc, char* argv[]) {
                  "e.g ./infer_model ./pptinypose_model_dir ./test.jpeg 0"
               << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with xpu."
               << std::endl;
     return -1;
   }
@@ -140,6 +174,8 @@ int main(int argc, char* argv[]) {
     GpuInfer(argv[1], argv[2]);
   } else if (std::atoi(argv[3]) == 2) {
     TrtInfer(argv[1], argv[2]);
+  } else if (std::atoi(argv[3]) == 3) {
+    TrtInfer(argv[1], argv[2]);
   }
   return 0;
 }

examples/vision/keypointdetection/tiny_pose/python/README.md

@@ -25,6 +25,8 @@ python pptinypose_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --imag
 python pptinypose_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --image hrnet_demo.jpg --device gpu
 # GPU上使用TensorRT推理 （注意：TensorRT推理第一次运行，有序列化模型的操作，有一定耗时，需要耐心等待）
 python pptinypose_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --image hrnet_demo.jpg --device gpu --use_trt True
+# XPU推理
+python pptinypose_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --image hrnet_demo.jpg --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/keypointdetection/tiny_pose/python/pptinypose_infer.py

@@ -17,7 +17,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="type of inference device, support 'cpu' or 'gpu'.")
+        help="type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--use_trt",
         type=ast.literal_eval,
@@ -32,6 +32,9 @@ def build_tinypose_option(args):
     if args.device.lower() == "gpu":
         option.use_gpu()
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+
     if args.use_trt:
         option.use_trt_backend()
         option.set_trt_input_shape("image", [1, 3, 256, 192])

examples/vision/matting/ppmatting/cpp/README.md

@@ -31,6 +31,8 @@ wget https://bj.bcebos.com/paddlehub/fastdeploy/matting_bgr.jpg
 ./infer_demo PP-Matting-512 matting_input.jpg matting_bgr.jpg 1
 # GPU上TensorRT推理
 ./infer_demo PP-Matting-512 matting_input.jpg matting_bgr.jpg 2
+# XPU推理
+./infer_demo PP-Matting-512 matting_input.jpg matting_bgr.jpg 3
 ```
 
 运行完成可视化结果如下图所示

examples/vision/matting/ppmatting/cpp/infer.cc

@@ -51,6 +51,37 @@ void CpuInfer(const std::string& model_dir, const std::string& image_file,
             << std::endl;
 }
 
+void XpuInfer(const std::string& model_dir, const std::string& image_file,
+              const std::string& background_file) {
+  auto model_file = model_dir + sep + "model.pdmodel";
+  auto params_file = model_dir + sep + "model.pdiparams";
+  auto config_file = model_dir + sep + "deploy.yaml";
+  auto option = fastdeploy::RuntimeOption();
+  option.UseXpu();
+  auto model = fastdeploy::vision::matting::PPMatting(model_file, params_file,
+                                                      config_file, option);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  cv::Mat bg = cv::imread(background_file);
+  fastdeploy::vision::MattingResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+  auto vis_im = fastdeploy::vision::VisMatting(im, res);
+  auto vis_im_with_bg =
+      fastdeploy::vision::SwapBackground(im, bg, res);
+  cv::imwrite("visualized_result.jpg", vis_im_with_bg);
+  cv::imwrite("visualized_result_fg.jpg", vis_im);
+  std::cout << "Visualized result save in ./visualized_result_replaced_bg.jpg "
+               "and ./visualized_result_fg.jpg"
+            << std::endl;
+}
+
 void GpuInfer(const std::string& model_dir, const std::string& image_file,
               const std::string& background_file) {
   auto model_file = model_dir + sep + "model.pdmodel";
@@ -125,7 +156,7 @@ int main(int argc, char* argv[]) {
            "e.g ./infer_model ./PP-Matting-512 ./test.jpg ./test_bg.jpg 0"
         << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend, 3: run with xpu."
               << std::endl;
     return -1;
   }
@@ -135,6 +166,8 @@ int main(int argc, char* argv[]) {
     GpuInfer(argv[1], argv[2], argv[3]);
   } else if (std::atoi(argv[4]) == 2) {
     TrtInfer(argv[1], argv[2], argv[3]);
+  } else if (std::atoi(argv[4]) == 3) {
+    XpuInfer(argv[1], argv[2], argv[3]);
   }
   return 0;
 }

examples/vision/matting/ppmatting/python/README.md

@@ -23,6 +23,8 @@ python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bg
 python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bgr.jpg --device gpu
 # GPU上使用TensorRT推理 （注意：TensorRT推理第一次运行，有序列化模型的操作，有一定耗时，需要耐心等待）
 python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bgr.jpg --device gpu --use_trt True
+# XPU推理
+python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bgr.jpg --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/matting/ppmatting/python/infer.py

@@ -21,7 +21,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        help="Type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--use_trt",
         type=ast.literal_eval,
@@ -40,6 +40,8 @@ def build_option(args):
         option.use_trt_backend()
         option.set_trt_input_shape("img", [1, 3, 512, 512])
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
     return option
 
 

examples/vision/ocr/PP-OCRv2/cpp/README.md

@@ -39,6 +39,10 @@ wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_
 ./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 1
 # GPU上TensorRT推理
 ./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 2
+# GPU上Paddle-TRT推理
+./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 3
+# XPU推理
+./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 4
 ```
 
 以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:  

examples/vision/ocr/PP-OCRv2/cpp/infer.cc

@@ -100,7 +100,7 @@ int main(int argc, char* argv[]) {
                  "./ppocr_keys_v1.txt ./12.jpg 0"
               << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with gpu and use Paddle-TRT; 4: run with xpu."
               << std::endl;
     return -1;
   }
@@ -120,6 +120,8 @@ int main(int argc, char* argv[]) {
     option.UseTrtBackend();
     option.EnablePaddleTrtCollectShape();
     option.EnablePaddleToTrt();
+  } else if (flag == 4) {
+    option.UseXpu();
   }
 
   std::string det_model_dir = argv[1];

examples/vision/ocr/PP-OCRv2/python/README.md

@@ -34,6 +34,8 @@ python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2
 python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu
 # GPU上使用TensorRT推理
 python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend trt
+# XPU推理
+python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/ocr/PP-OCRv2/python/infer.py

@@ -41,7 +41,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        help="Type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--backend",
         type=str,
@@ -68,6 +68,10 @@ def build_option(args):
 
     option.set_cpu_thread_num(args.cpu_thread_num)
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+        return option
+
     if args.backend.lower() == "trt":
         assert args.device.lower(
         ) == "gpu", "TensorRT backend require inference on device GPU."

examples/vision/ocr/PP-OCRv3/cpp/README.md

@@ -39,6 +39,10 @@ wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_
 ./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 1
 # GPU上TensorRT推理
 ./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 2
+# GPU上Paddle-TRT推理
+./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 3
+# XPU推理
+./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 4
 ```
 
 以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:  

examples/vision/ocr/PP-OCRv3/cpp/infer.cc

@@ -101,7 +101,7 @@ int main(int argc, char* argv[]) {
                  "./ppocr_keys_v1.txt ./12.jpg 0"
               << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with gpu and use Paddle-TRT; 4: run with xpu."
               << std::endl;
     return -1;
   }
@@ -121,6 +121,8 @@ int main(int argc, char* argv[]) {
     option.UseTrtBackend();
     option.EnablePaddleTrtCollectShape();
     option.EnablePaddleToTrt();
+  } else if (flag == 4) {
+    option.UseXpu();
   }
 
   std::string det_model_dir = argv[1];

examples/vision/ocr/PP-OCRv3/python/README.md

@@ -33,6 +33,8 @@ python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2
 python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu
 # GPU上使用TensorRT推理
 python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend trt
+# XPU推理
+python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/ocr/PP-OCRv3/python/infer.py

@@ -41,7 +41,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        help="Type of inference device, support 'cpu', 'xpu' or 'gpu'.")
     parser.add_argument(
         "--backend",
         type=str,
@@ -68,6 +68,10 @@ def build_option(args):
 
     option.set_cpu_thread_num(args.cpu_thread_num)
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+        return option
+
     if args.backend.lower() == "trt":
         assert args.device.lower(
         ) == "gpu", "TensorRT backend require inference on device GPU."

examples/vision/segmentation/paddleseg/cpp/README.md

@@ -32,6 +32,8 @@ wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
 ./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 1
 # GPU上TensorRT推理
 ./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 2
+# XPU推理
+./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 3
 ```
 
 运行完成可视化结果如下图所示

examples/vision/segmentation/paddleseg/cpp/infer.cc

@@ -48,6 +48,34 @@ void CpuInfer(const std::string& model_dir, const std::string& image_file) {
   std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
 }
 
+void XpuInfer(const std::string& model_dir, const std::string& image_file) {
+  auto model_file = model_dir + sep + "model.pdmodel";
+  auto params_file = model_dir + sep + "model.pdiparams";
+  auto config_file = model_dir + sep + "deploy.yaml";
+  auto option = fastdeploy::RuntimeOption();
+  option.UseXpu();
+  auto model = fastdeploy::vision::segmentation::PaddleSegModel(
+      model_file, params_file, config_file, option);
+
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+
+  fastdeploy::vision::SegmentationResult res;
+  if (!model.Predict(im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+
+  std::cout << res.Str() << std::endl;
+  auto vis_im = fastdeploy::vision::VisSegmentation(im, res, 0.5);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
+}
+
 void GpuInfer(const std::string& model_dir, const std::string& image_file) {
   auto model_file = model_dir + sep + "model.pdmodel";
   auto params_file = model_dir + sep + "model.pdiparams";
@@ -114,7 +142,7 @@ int main(int argc, char* argv[]) {
            "e.g ./infer_model ./ppseg_model_dir ./test.jpeg 0"
         << std::endl;
     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
-                 "with gpu; 2: run with gpu and use tensorrt backend."
+                 "with gpu; 2: run with gpu and use tensorrt backend; 3: run with xpu."
               << std::endl;
     return -1;
   }
@@ -125,6 +153,8 @@ int main(int argc, char* argv[]) {
     GpuInfer(argv[1], argv[2]);
   } else if (std::atoi(argv[3]) == 2) {
     TrtInfer(argv[1], argv[2]);
+  } else if (std::atoi(argv[3]) == 3) {
+    XpuInfer(argv[1], argv[2]);
   }
   return 0;
 }

examples/vision/segmentation/paddleseg/python/README.md

@@ -25,6 +25,8 @@ python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_
 python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device gpu
 # GPU上使用TensorRT推理 （注意：TensorRT推理第一次运行，有序列化模型的操作，有一定耗时，需要耐心等待）
 python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device gpu --use_trt True
+# XPU推理
+python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device xpu
 ```
 
 运行完成可视化结果如下图所示

examples/vision/segmentation/paddleseg/python/infer.py

@@ -15,7 +15,7 @@ def parse_arguments():
         "--device",
         type=str,
         default='cpu',
-        help="Type of inference device, support 'cpu' or 'gpu'.")
+        help="Type of inference device, support 'xpu', 'cpu' or 'gpu'.")
     parser.add_argument(
         "--use_trt",
         type=ast.literal_eval,
@@ -30,6 +30,9 @@ def build_option(args):
     if args.device.lower() == "gpu":
         option.use_gpu()
 
+    if args.device.lower() == "xpu":
+        option.use_xpu()
+
     if args.use_trt:
         option.use_trt_backend()
         option.set_trt_input_shape("x", [1, 3, 256, 256], [1, 3, 1024, 1024],

fastdeploy/vision/faceid/contrib/insightface_rec.cc

@@ -32,6 +32,7 @@ InsightFaceRecognitionModel::InsightFaceRecognitionModel(
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::LITE};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_xpu_backends = {Backend::LITE};
   }
   runtime_option = custom_option;
   runtime_option.model_format = model_format;

fastdeploy/vision/keypointdet/pptinypose/pptinypose.cc

@@ -18,6 +18,7 @@ PPTinyPose::PPTinyPose(const std::string& model_file,
   config_file_ = config_file;
   valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO, Backend::LITE};
   valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+  valid_xpu_backends = {Backend::LITE};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
   runtime_option.model_file = model_file;

fastdeploy/vision/matting/ppmatting/ppmatting.cc

@@ -28,6 +28,7 @@ PPMatting::PPMatting(const std::string& model_file,
   config_file_ = config_file;
   valid_cpu_backends = {Backend::ORT, Backend::PDINFER, Backend::LITE};
   valid_gpu_backends = {Backend::PDINFER, Backend::TRT};
+  valid_xpu_backends = {Backend::LITE};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
   runtime_option.model_file = model_file;

fastdeploy/vision/ocr/ppocr/classifier.cc

@@ -32,6 +32,7 @@ Classifier::Classifier(const std::string& model_file,
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO, Backend::LITE};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_xpu_backends = {Backend::LITE};
     valid_ascend_backends = {Backend::LITE};
   }
   runtime_option = custom_option;

fastdeploy/vision/ocr/ppocr/dbdetector.cc

@@ -32,6 +32,7 @@ DBDetector::DBDetector(const std::string& model_file,
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO, Backend::LITE};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_xpu_backends = {Backend::LITE};
     valid_ascend_backends = {Backend::LITE};
   }
 

fastdeploy/vision/ocr/ppocr/recognizer.cc

@@ -34,6 +34,7 @@ Recognizer::Recognizer(const std::string& model_file,
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::OPENVINO, Backend::LITE};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_xpu_backends = {Backend::LITE};
     valid_ascend_backends = {Backend::LITE}; 
   }
 

fastdeploy/vision/segmentation/ppseg/model.cc

@@ -29,6 +29,7 @@ PaddleSegModel::PaddleSegModel(const std::string& model_file,
   valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
   valid_rknpu_backends = {Backend::RKNPU2};
   valid_timvx_backends = {Backend::LITE};
+  valid_xpu_backends = {Backend::LITE};
   valid_ascend_backends = {Backend::LITE};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
@@ -89,4 +90,4 @@ bool PaddleSegModel::BatchPredict(const std::vector<cv::Mat>& imgs,
 }
 }  // namespace segmentation
 }  // namespace vision
-}  // namespace fastdeploy
+}  // namespace fastdeploy