diff --git a/benchmark/benchmark_ppcls.py b/benchmark/benchmark_ppcls.py
new file mode 100644
index 000000000..75381ceeb
--- /dev/null
+++ b/benchmark/benchmark_ppcls.py
@@ -0,0 +1,170 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import fastdeploy as fd
+import cv2
+import os
+import numpy as np
+import pynvml
+import psutil
+import GPUtil
+import time
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", required=True, help="Path of PaddleClas model.")
+    parser.add_argument(
+        "--image", type=str, required=False, help="Path of test image file.")
+    parser.add_argument(
+        "--cpu_num_thread",
+        type=int,
+        default=8,
+        help="default number of cpu thread.")
+    parser.add_argument(
+        "--device_id", type=int, default=0, help="device(gpu) id")
+    parser.add_argument(
+        "--iter_num",
+        required=True,
+        type=int,
+        default=300,
+        help="number of iterations for computing performace.")
+    parser.add_argument(
+        "--device",
+        default="cpu",
+        help="Type of inference device, support 'cpu' or 'gpu'.")
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="ort",
+        help="inference backend, ort, ov, trt, paddle.")
+    parser.add_argument(
+        "--enable_trt_fp16",
+        type=bool,
+        default=False,
+        help="whether enable fp16 in trt backend")
+    args = parser.parse_args()
+    return args
+
+
+def build_option(args):
+    option = fd.RuntimeOption()
+    device = args.device
+    backend = args.backend
+    option.set_cpu_thread_num(args.cpu_num_thread)
+    if device == "gpu":
+        option.use_gpu(args.device_id)
+
+    if backend == "trt":
+        assert device == "gpu", "the trt backend need device==gpu"
+        option.use_trt_backend()
+        if args.enable_trt_fp16:
+            option.enable_trt_fp16()
+    elif backend == "ov":
+        assert device == "cpu", "the openvino backend need device==cpu"
+        option.use_openvino_backend()
+    elif backend == "paddle":
+        option.use_paddle_backend()
+    elif backend == "ort":
+        option.use_ort_backend()
+    else:
+        print("%s is an unsupported backend" % backend)
+
+    return option
+
+
+def get_current_memory_mb(gpu_id=None):
+    pid = os.getpid()
+    p = psutil.Process(pid)
+    info = p.memory_full_info()
+    cpu_mem = info.uss / 1024. / 1024.
+    gpu_mem = 0
+    if gpu_id is not None:
+        pynvml.nvmlInit()
+        handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+        meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
+        gpu_mem = meminfo.used / 1024. / 1024.
+    return cpu_mem, gpu_mem
+
+
+def get_current_gputil(gpu_id):
+    GPUs = GPUtil.getGPUs()
+    gpu_load = GPUs[gpu_id].load
+    return gpu_load
+
+
+if __name__ == '__main__':
+
+    args = parse_arguments()
+    option = build_option(args)
+    model_file = os.path.join(args.model, "inference.pdmodel")
+    params_file = os.path.join(args.model, "inference.pdiparams")
+    config_file = os.path.join(args.model, "inference_cls.yaml")
+
+    gpu_id = args.device_id
+    end2end_statis = list()
+    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
+    if args.device == "cpu":
+        file_path = args.model + "_model_" + args.backend + "_" + \
+            args.device + "_" + str(args.cpu_num_thread) + ".txt"
+    else:
+        if args.enable_trt_fp16:
+            file_path = args.model + "_model_" + \
+                args.backend + "_fp16_" + args.device + ".txt"
+        else:
+            file_path = args.model + "_model_" + args.backend + "_" + args.device + ".txt"
+    f = open(file_path, "w")
+    f.writelines("===={}====: \n".format(file_path.split("/")[1][:-4]))
+
+    try:
+        model = fd.vision.classification.PaddleClasModel(
+            model_file, params_file, config_file, runtime_option=option)
+        model.enable_record_time_of_runtime()
+        for i in range(args.iter_num):
+            im = cv2.imread(args.image)
+            start = time.time()
+            result = model.predict(im)
+            end2end_statis.append(time.time() - start)
+            gpu_util += get_current_gputil(gpu_id)
+            cm, gm = get_current_memory_mb(gpu_id)
+            cpu_mem += cm
+            gpu_mem += gm
+
+        runtime_statis = model.print_statis_info_of_runtime()
+
+        warmup_iter = args.iter_num // 5
+        repeat_iter = args.iter_num - warmup_iter
+        end2end_statis = end2end_statis[warmup_iter:]
+
+        dump_result = dict()
+        dump_result["runtime"] = runtime_statis["avg_time"] * 1000
+        dump_result["end2end"] = np.mean(end2end_statis) * 1000
+        dump_result["cpu_rss_mb"] = cpu_mem / repeat_iter
+        dump_result["gpu_rss_mb"] = gpu_mem / repeat_iter
+        dump_result["gpu_util"] = gpu_util / repeat_iter
+
+        f.writelines("Runtime(ms): {} \n".format(str(dump_result["runtime"])))
+        f.writelines("End2End(ms): {} \n".format(str(dump_result["end2end"])))
+        f.writelines("cpu_rss_mb: {} \n".format(
+            str(dump_result["cpu_rss_mb"])))
+        f.writelines("gpu_rss_mb: {} \n".format(
+            str(dump_result["gpu_rss_mb"])))
+        f.writelines("gpu_util: {} \n".format(str(dump_result["gpu_util"])))
+    except:
+        f.writelines("!!!!!Infer Failed\n")
+
+    f.close()
diff --git a/benchmark/benchmark_ppdet.py b/benchmark/benchmark_ppdet.py
new file mode 100644
index 000000000..71fb665c4
--- /dev/null
+++ b/benchmark/benchmark_ppdet.py
@@ -0,0 +1,190 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import fastdeploy as fd
+import cv2
+import os
+import numpy as np
+import datetime
+import json
+import pynvml
+import psutil
+import GPUtil
+import time
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", required=True, help="Path of PaddleDetection model.")
+    parser.add_argument(
+        "--image", type=str, required=False, help="Path of test image file.")
+    parser.add_argument(
+        "--cpu_num_thread",
+        type=int,
+        default=8,
+        help="default number of cpu thread.")
+    parser.add_argument(
+        "--device_id", type=int, default=0, help="device(gpu) id")
+    parser.add_argument(
+        "--iter_num",
+        required=True,
+        type=int,
+        default=300,
+        help="number of iterations for computing performace.")
+    parser.add_argument(
+        "--device",
+        default="cpu",
+        help="Type of inference device, support 'cpu' or 'gpu'.")
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="ort",
+        help="inference backend, ort, ov, trt, paddle.")
+    parser.add_argument(
+        "--enable_trt_fp16",
+        type=bool,
+        default=False,
+        help="whether enable fp16 in trt backend")
+    args = parser.parse_args()
+    return args
+
+
+def build_option(args):
+    option = fd.RuntimeOption()
+    device = args.device
+    backend = args.backend
+    option.set_cpu_thread_num(args.cpu_num_thread)
+    if device == "gpu":
+        option.use_gpu(args.device_id)
+
+    if backend == "trt":
+        assert device == "gpu", "the trt backend need device==gpu"
+        option.use_trt_backend()
+        if args.enable_trt_fp16:
+            option.enable_trt_fp16()
+    elif backend == "ov":
+        assert device == "cpu", "the openvino backend need device==cpu"
+        option.use_openvino_backend()
+    elif backend == "paddle":
+        option.use_paddle_backend()
+    elif backend == "ort":
+        option.use_ort_backend()
+    else:
+        print("%s is an unsupported backend" % backend)
+
+    return option
+
+
+def get_current_memory_mb(gpu_id=None):
+    pid = os.getpid()
+    p = psutil.Process(pid)
+    info = p.memory_full_info()
+    cpu_mem = info.uss / 1024. / 1024.
+    gpu_mem = 0
+    if gpu_id is not None:
+        pynvml.nvmlInit()
+        handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+        meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
+        gpu_mem = meminfo.used / 1024. / 1024.
+    return cpu_mem, gpu_mem
+
+
+def get_current_gputil(gpu_id):
+    GPUs = GPUtil.getGPUs()
+    gpu_load = GPUs[gpu_id].load
+    return gpu_load
+
+
+if __name__ == '__main__':
+
+    args = parse_arguments()
+    option = build_option(args)
+    model_file = os.path.join(args.model, "model.pdmodel")
+    params_file = os.path.join(args.model, "model.pdiparams")
+    config_file = os.path.join(args.model, "infer_cfg.yml")
+
+    gpu_id = args.device_id
+    end2end_statis = list()
+    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
+    if args.device == "cpu":
+        file_path = args.model + "_model_" + args.backend + "_" + \
+            args.device + "_" + str(args.cpu_num_thread) + ".txt"
+    else:
+        if args.enable_trt_fp16:
+            file_path = args.model + "_model_" + args.backend + "_fp16_" + args.device + ".txt"
+        else:
+            file_path = args.model + "_model_" + args.backend + "_" + args.device + ".txt"
+    f = open(file_path, "w")
+    f.writelines("===={}====: \n".format(file_path.split("/")[1][:-4]))
+
+    try:
+        if "ppyoloe" in args.model:
+            model = fd.vision.detection.PPYOLOE(
+                model_file, params_file, config_file, runtime_option=option)
+        elif "picodet" in args.model:
+            model = fd.vision.detection.PicoDet(
+                model_file, params_file, config_file, runtime_option=option)
+        elif "yolox" in args.model:
+            model = fd.vision.detection.PaddleYOLOX(
+                model_file, params_file, config_file, runtime_option=option)
+        elif "yolov3" in args.model:
+            model = fd.vision.detection.YOLOv3(
+                model_file, params_file, config_file, runtime_option=option)
+        elif "ppyolo_r50vd_dcn_1x_coco" in args.model or "ppyolov2_r101vd_dcn_365e_coco" in args.model:
+            model = fd.vision.detection.PPYOLO(
+                model_file, params_file, config_file, runtime_option=option)
+        elif "faster_rcnn" in args.model:
+            model = fd.vision.detection.FasterRCNN(
+                model_file, params_file, config_file, runtime_option=option)
+        else:
+            raise Exception("model {} not support now in ppdet series".format(
+                args.model))
+        model.enable_record_time_of_runtime()
+        for i in range(args.iter_num):
+            im = cv2.imread(args.image)
+            start = time.time()
+            result = model.predict(im)
+            end2end_statis.append(time.time() - start)
+            gpu_util += get_current_gputil(gpu_id)
+            cm, gm = get_current_memory_mb(gpu_id)
+            cpu_mem += cm
+            gpu_mem += gm
+
+        runtime_statis = model.print_statis_info_of_runtime()
+
+        warmup_iter = args.iter_num // 5
+        repeat_iter = args.iter_num - warmup_iter
+        end2end_statis = end2end_statis[warmup_iter:]
+
+        dump_result = dict()
+        dump_result["runtime"] = runtime_statis["avg_time"] * 1000
+        dump_result["end2end"] = np.mean(end2end_statis) * 1000
+        dump_result["cpu_rss_mb"] = cpu_mem / repeat_iter
+        dump_result["gpu_rss_mb"] = gpu_mem / repeat_iter
+        dump_result["gpu_util"] = gpu_util / repeat_iter
+
+        f.writelines("Runtime(ms): {} \n".format(str(dump_result["runtime"])))
+        f.writelines("End2End(ms): {} \n".format(str(dump_result["end2end"])))
+        f.writelines("cpu_rss_mb: {} \n".format(
+            str(dump_result["cpu_rss_mb"])))
+        f.writelines("gpu_rss_mb: {} \n".format(
+            str(dump_result["gpu_rss_mb"])))
+        f.writelines("gpu_util: {} \n".format(str(dump_result["gpu_util"])))
+    except:
+        f.writelines("!!!!!Infer Failed\n")
+
+    f.close()
diff --git a/benchmark/benchmark_ppseg.py b/benchmark/benchmark_ppseg.py
new file mode 100644
index 000000000..889264c9f
--- /dev/null
+++ b/benchmark/benchmark_ppseg.py
@@ -0,0 +1,169 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import fastdeploy as fd
+import cv2
+import os
+import numpy as np
+import pynvml
+import psutil
+import GPUtil
+import time
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", required=True, help="Path of PaddleSeg model.")
+    parser.add_argument(
+        "--image", type=str, required=False, help="Path of test image file.")
+    parser.add_argument(
+        "--cpu_num_thread",
+        type=int,
+        default=8,
+        help="default number of cpu thread.")
+    parser.add_argument(
+        "--device_id", type=int, default=0, help="device(gpu) id")
+    parser.add_argument(
+        "--iter_num",
+        required=True,
+        type=int,
+        default=300,
+        help="number of iterations for computing performace.")
+    parser.add_argument(
+        "--device",
+        default="cpu",
+        help="Type of inference device, support 'cpu' or 'gpu'.")
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="ort",
+        help="inference backend, ort, ov, trt, paddle.")
+    parser.add_argument(
+        "--enable_trt_fp16",
+        type=bool,
+        default=False,
+        help="whether enable fp16 in trt backend")
+    args = parser.parse_args()
+    return args
+
+
+def build_option(args):
+    option = fd.RuntimeOption()
+    device = args.device
+    backend = args.backend
+    option.set_cpu_thread_num(args.cpu_num_thread)
+    if device == "gpu":
+        option.use_gpu(args.device_id)
+
+    if backend == "trt":
+        assert device == "gpu", "the trt backend need device==gpu"
+        option.use_trt_backend()
+        if args.enable_trt_fp16:
+            option.enable_trt_fp16()
+    elif backend == "ov":
+        assert device == "cpu", "the openvino backend need device==cpu"
+        option.use_openvino_backend()
+    elif backend == "paddle":
+        option.use_paddle_backend()
+    elif backend == "ort":
+        option.use_ort_backend()
+    else:
+        print("%s is an unsupported backend" % backend)
+
+    return option
+
+
+def get_current_memory_mb(gpu_id=None):
+    pid = os.getpid()
+    p = psutil.Process(pid)
+    info = p.memory_full_info()
+    cpu_mem = info.uss / 1024. / 1024.
+    gpu_mem = 0
+    if gpu_id is not None:
+        pynvml.nvmlInit()
+        handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+        meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
+        gpu_mem = meminfo.used / 1024. / 1024.
+    return cpu_mem, gpu_mem
+
+
+def get_current_gputil(gpu_id):
+    GPUs = GPUtil.getGPUs()
+    gpu_load = GPUs[gpu_id].load
+    return gpu_load
+
+
+if __name__ == '__main__':
+
+    args = parse_arguments()
+    option = build_option(args)
+    model_file = os.path.join(args.model, "model.pdmodel")
+    params_file = os.path.join(args.model, "model.pdiparams")
+    config_file = os.path.join(args.model, "deploy.yaml")
+
+    gpu_id = args.device_id
+    end2end_statis = list()
+    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
+    if args.device == "cpu":
+        file_path = args.model + "_model_" + args.backend + "_" + \
+            args.device + "_" + str(args.cpu_num_thread) + ".txt"
+    else:
+        if args.enable_trt_fp16:
+            file_path = args.model + "_model_" + args.backend + "_fp16_" + args.device + ".txt"
+        else:
+            file_path = args.model + "_model_" + args.backend + "_" + args.device + ".txt"
+    f = open(file_path, "w")
+    f.writelines("===={}====: \n".format(file_path.split("/")[1][:-4]))
+
+    try:
+        model = fd.vision.segmentation.PaddleSegModel(
+            model_file, params_file, config_file, runtime_option=option)
+        model.enable_record_time_of_runtime()
+        for i in range(args.iter_num):
+            im = cv2.imread(args.image)
+            start = time.time()
+            result = model.predict(im)
+            end2end_statis.append(time.time() - start)
+            gpu_util += get_current_gputil(gpu_id)
+            cm, gm = get_current_memory_mb(gpu_id)
+            cpu_mem += cm
+            gpu_mem += gm
+
+        runtime_statis = model.print_statis_info_of_runtime()
+
+        warmup_iter = args.iter_num // 5
+        repeat_iter = args.iter_num - warmup_iter
+        end2end_statis = end2end_statis[warmup_iter:]
+
+        dump_result = dict()
+        dump_result["runtime"] = runtime_statis["avg_time"] * 1000
+        dump_result["end2end"] = np.mean(end2end_statis) * 1000
+        dump_result["cpu_rss_mb"] = cpu_mem / repeat_iter
+        dump_result["gpu_rss_mb"] = gpu_mem / repeat_iter
+        dump_result["gpu_util"] = gpu_util / repeat_iter
+
+        f.writelines("Runtime(ms): {} \n".format(str(dump_result["runtime"])))
+        f.writelines("End2End(ms): {} \n".format(str(dump_result["end2end"])))
+        f.writelines("cpu_rss_mb: {} \n".format(
+            str(dump_result["cpu_rss_mb"])))
+        f.writelines("gpu_rss_mb: {} \n".format(
+            str(dump_result["gpu_rss_mb"])))
+        f.writelines("gpu_util: {} \n".format(str(dump_result["gpu_util"])))
+    except:
+        f.writelines("!!!!!Infer Failed\n")
+
+    f.close()
diff --git a/benchmark/benchmark_yolo.py b/benchmark/benchmark_yolo.py
new file mode 100644
index 000000000..6c7fec6e4
--- /dev/null
+++ b/benchmark/benchmark_yolo.py
@@ -0,0 +1,183 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import fastdeploy as fd
+import cv2
+import os
+import numpy as np
+import datetime
+import json
+import pynvml
+import psutil
+import GPUtil
+import time
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", required=True, help="Path of Yolo onnx model.")
+    parser.add_argument(
+        "--image", type=str, required=False, help="Path of test image file.")
+    parser.add_argument(
+        "--cpu_num_thread",
+        type=int,
+        default=8,
+        help="default number of cpu thread.")
+    parser.add_argument(
+        "--device_id", type=int, default=0, help="device(gpu) id")
+    parser.add_argument(
+        "--iter_num",
+        required=True,
+        type=int,
+        default=300,
+        help="number of iterations for computing performace.")
+    parser.add_argument(
+        "--device",
+        default="cpu",
+        help="Type of inference device, support 'cpu' or 'gpu'.")
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="ort",
+        help="inference backend, ort, ov, trt, paddle.")
+    parser.add_argument(
+        "--enable_trt_fp16",
+        type=bool,
+        default=False,
+        help="whether enable fp16 in trt backend")
+    args = parser.parse_args()
+    return args
+
+
+def build_option(args):
+    option = fd.RuntimeOption()
+    device = args.device
+    backend = args.backend
+    option.set_cpu_thread_num(args.cpu_num_thread)
+    if device == "gpu":
+        option.use_gpu(args.device_id)
+
+    if backend == "trt":
+        assert device == "gpu", "the trt backend need device==gpu"
+        option.use_trt_backend()
+        if args.enable_trt_fp16:
+            option.enable_trt_fp16()
+    elif backend == "ov":
+        assert device == "cpu", "the openvino backend need device==cpu"
+        option.use_openvino_backend()
+    elif backend == "paddle":
+        option.use_paddle_backend()
+    elif backend == "ort":
+        option.use_ort_backend()
+    else:
+        print("%s is an unsupported backend" % backend)
+
+    return option
+
+
+def get_current_memory_mb(gpu_id=None):
+    pid = os.getpid()
+    p = psutil.Process(pid)
+    info = p.memory_full_info()
+    cpu_mem = info.uss / 1024. / 1024.
+    gpu_mem = 0
+    if gpu_id is not None:
+        pynvml.nvmlInit()
+        handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+        meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
+        gpu_mem = meminfo.used / 1024. / 1024.
+    return cpu_mem, gpu_mem
+
+
+def get_current_gputil(gpu_id):
+    GPUs = GPUtil.getGPUs()
+    gpu_load = GPUs[gpu_id].load
+    return gpu_load
+
+
+if __name__ == '__main__':
+
+    args = parse_arguments()
+    option = build_option(args)
+    model_file = args.model
+
+    gpu_id = args.device_id
+    end2end_statis = list()
+    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
+    if args.device == "cpu":
+        file_path = args.model + "_model_" + args.backend + "_" + \
+            args.device + "_" + str(args.cpu_num_thread) + ".txt"
+    else:
+        if args.enable_trt_fp16:
+            file_path = args.model + "_model_" + args.backend + "_fp16_" + args.device + ".txt"
+        else:
+            file_path = args.model + "_model_" + args.backend + "_" + args.device + ".txt"
+    f = open(file_path, "w")
+    f.writelines("===={}====: \n".format(file_path.split("/")[1][:-4]))
+
+    try:
+        if "yolox" in model_file:
+            model = fd.vision.detection.YOLOX(
+                model_file, runtime_option=option)
+        elif "yolov5" in model_file:
+            model = fd.vision.detection.YOLOv5(
+                model_file, runtime_option=option)
+        elif "yolov6" in model_file:
+            model = fd.vision.detection.YOLOv6(
+                model_file, runtime_option=option)
+        elif "yolov7" in model_file:
+            model = fd.vision.detection.YOLOv7(
+                model_file, runtime_option=option)
+        else:
+            raise Exception("model {} not support now in yolo series".format(
+                args.model))
+        model.enable_record_time_of_runtime()
+
+        for i in range(args.iter_num):
+            im = cv2.imread(args.image)
+            start = time.time()
+            result = model.predict(im)
+            end2end_statis.append(time.time() - start)
+            gpu_util += get_current_gputil(gpu_id)
+            cm, gm = get_current_memory_mb(gpu_id)
+            cpu_mem += cm
+            gpu_mem += gm
+
+        runtime_statis = model.print_statis_info_of_runtime()
+
+        warmup_iter = args.iter_num // 5
+        repeat_iter = args.iter_num - warmup_iter
+        end2end_statis = end2end_statis[warmup_iter:]
+
+        dump_result = dict()
+        dump_result["runtime"] = runtime_statis["avg_time"] * 1000
+        dump_result["end2end"] = np.mean(end2end_statis) * 1000
+        dump_result["cpu_rss_mb"] = cpu_mem / repeat_iter
+        dump_result["gpu_rss_mb"] = gpu_mem / repeat_iter
+        dump_result["gpu_util"] = gpu_util / repeat_iter
+
+        f.writelines("Runtime(ms): {} \n".format(str(dump_result["runtime"])))
+        f.writelines("End2End(ms): {} \n".format(str(dump_result["end2end"])))
+        f.writelines("cpu_rss_mb: {} \n".format(
+            str(dump_result["cpu_rss_mb"])))
+        f.writelines("gpu_rss_mb: {} \n".format(
+            str(dump_result["gpu_rss_mb"])))
+        f.writelines("gpu_util: {} \n".format(str(dump_result["gpu_util"])))
+    except:
+        f.writelines("!!!!!Infer Failed\n")
+
+    f.close()
diff --git a/benchmark/convert_info.py b/benchmark/convert_info.py
new file mode 100644
index 000000000..893ffecbe
--- /dev/null
+++ b/benchmark/convert_info.py
@@ -0,0 +1,155 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import argparse
+
+parser = argparse.ArgumentParser(description='manual to this script')
+parser.add_argument('--txt_path', type=str, default='result.txt')
+parser.add_argument('--domain', type=str, default='ppcls')
+args = parser.parse_args()
+txt_path = args.txt_path
+domain = args.domain
+
+f1 = open(txt_path, "r")
+lines = f1.readlines()
+line_nums = len(lines)
+ort_cpu_thread1 = dict()
+ort_cpu_thread8 = dict()
+ort_gpu = dict()
+ov_cpu_thread1 = dict()
+ov_cpu_thread8 = dict()
+paddle_cpu_thread1 = dict()
+paddle_cpu_thread8 = dict()
+paddle_gpu = dict()
+trt_gpu = dict()
+trt_gpu_fp16 = dict()
+model_name_set = set()
+
+for i in range(line_nums):
+    if "====" in lines[i]:
+        model_name = lines[i].strip().split("_model")[0][4:]
+        model_name_set.add(model_name)
+        runtime = "-"
+        end2end = "-"
+        if "Runtime(ms)" in lines[i + 1]:
+            runtime_ori = lines[i + 1].split(": ")[1]
+            # two decimal places
+            runtime_list = runtime_ori.split(".")
+            runtime = runtime_list[0] + "." + runtime_list[1][:2]
+        if "End2End(ms)" in lines[i + 2]:
+            end2end_ori = lines[i + 2].split(": ")[1]
+            # two decimal places
+            end2end_list = end2end_ori.split(".")
+            end2end = end2end_list[0] + "." + end2end_list[1][:2]
+        if "cpu_rss_mb" in lines[i + 3]:
+            cpu_rss_mb_ori = lines[i + 3].split(": ")[1]
+            # two decimal places
+            cpu_rss_mb_list = cpu_rss_mb_ori.split(".")
+            cpu_rss_mb = cpu_rss_mb_list[0] + "." + cpu_rss_mb_list[1][:2]
+        if "gpu_rss_mb" in lines[i + 4]:
+            gpu_rss_mb_ori = lines[i + 4].split(": ")[1]
+            # two decimal places
+            gpu_rss_mb_list = gpu_rss_mb_ori.split(".")
+            gpu_rss_mb = gpu_rss_mb_list[0] + "." + gpu_rss_mb_list[1][:2]
+        if "ort_cpu_1" in lines[i]:
+            ort_cpu_thread1[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "ort_cpu_8" in lines[i]:
+            ort_cpu_thread8[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "ort_gpu" in lines[i]:
+            ort_gpu[model_name] = runtime + "\t" + end2end + "\t" + gpu_rss_mb
+        elif "ov_cpu_1" in lines[i]:
+            ov_cpu_thread1[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "ov_cpu_8" in lines[i]:
+            ov_cpu_thread8[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "paddle_cpu_1" in lines[i]:
+            paddle_cpu_thread1[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "paddle_cpu_8" in lines[i]:
+            paddle_cpu_thread8[
+                model_name] = runtime + "\t" + end2end + "\t" + cpu_rss_mb
+        elif "paddle_gpu" in lines[i]:
+            paddle_gpu[
+                model_name] = runtime + "\t" + end2end + "\t" + gpu_rss_mb
+        elif "trt_gpu" in lines[i]:
+            trt_gpu[model_name] = runtime + "\t" + end2end + "\t" + gpu_rss_mb
+        elif "trt_fp16_gpu" in lines[i]:
+            trt_gpu_fp16[
+                model_name] = runtime + "\t" + end2end + "\t" + gpu_rss_mb
+
+f2 = open("struct_cpu_" + domain + ".txt", "w")
+f2.writelines(
+    "model_name\tthread_nums\tort_run\tort_end2end\tcpu_rss_mb\tov_run\tov_end2end\tcpu_rss_mb\tpaddle_run\tpaddle_end2end\tcpu_rss_mb\n"
+)
+for model_name in model_name_set:
+    lines1 = model_name + '\t1\t'
+    lines2 = model_name + '\t8\t'
+    if model_name in ort_cpu_thread1 and ort_cpu_thread1[model_name] != "":
+        lines1 += ort_cpu_thread1[model_name] + '\t'
+    else:
+        lines1 += "-\t-\t-\t"
+    if model_name in ov_cpu_thread1 and ov_cpu_thread1[model_name] != "":
+        lines1 += ov_cpu_thread1[model_name] + '\t'
+    else:
+        lines1 += "-\t-\t-\t"
+    if model_name in paddle_cpu_thread1 and paddle_cpu_thread1[
+            model_name] != "":
+        lines1 += paddle_cpu_thread1[model_name] + '\n'
+    else:
+        lines1 += "-\t-\t-\n"
+    f2.writelines(lines1)
+    if model_name in ort_cpu_thread8 and ort_cpu_thread8[model_name] != "":
+        lines2 += ort_cpu_thread8[model_name] + '\t'
+    else:
+        lines2 += "-\t-\t-\t"
+    if model_name in ov_cpu_thread8 and ov_cpu_thread8[model_name] != "":
+        lines2 += ov_cpu_thread8[model_name] + '\t'
+    else:
+        lines2 += "-\t-\t-\t"
+    if model_name in paddle_cpu_thread8 and paddle_cpu_thread8[
+            model_name] != "":
+        lines2 += paddle_cpu_thread8[model_name] + '\n'
+    else:
+        lines2 += "-\t-\t-\n"
+    f2.writelines(lines2)
+f2.close()
+
+f3 = open("struct_gpu_" + domain + ".txt", "w")
+f3.writelines(
+    "model_name\tort_run\tort_end2end\tgpu_rss_mb\tpaddle_run\tpaddle_end2end\tgpu_rss_mb\ttrt_run\ttrt_end2end\tgpu_rss_mb\ttrt_fp16_run\ttrt_fp16_end2end\tgpu_rss_mb\n"
+)
+for model_name in model_name_set:
+    lines1 = model_name + '\t'
+    if model_name in ort_gpu and ort_gpu[model_name] != "":
+        lines1 += ort_gpu[model_name] + '\t'
+    else:
+        lines1 += "-\t-\t-\t"
+    if model_name in paddle_gpu and paddle_gpu[model_name] != "":
+        lines1 += paddle_gpu[model_name] + '\t'
+    else:
+        lines1 += "-\t-\t-\t"
+    if model_name in trt_gpu and trt_gpu[model_name] != "":
+        lines1 += trt_gpu[model_name] + '\t'
+    else:
+        lines1 += "-\t-\t-\t"
+    if model_name in trt_gpu_fp16 and trt_gpu_fp16[model_name] != "":
+        lines1 += trt_gpu_fp16[model_name] + '\n'
+    else:
+        lines1 += "-\t-\t-\n"
+    f3.writelines(lines1)
+f3.close()
diff --git a/benchmark/requirements.txt b/benchmark/requirements.txt
new file mode 100644
index 000000000..9f1d255ea
--- /dev/null
+++ b/benchmark/requirements.txt
@@ -0,0 +1,6 @@
+numpy
+pynvml
+psutil
+GPUtil
+time
+numpy
diff --git a/benchmark/run_benchmark_ppcls.sh b/benchmark/run_benchmark_ppcls.sh
new file mode 100644
index 000000000..71e33a931
--- /dev/null
+++ b/benchmark/run_benchmark_ppcls.sh
@@ -0,0 +1,33 @@
+echo "[FastDeploy]    Running PPcls benchmark..."
+
+num_of_models=$(ls -d ppcls_model/* | wc -l)
+
+counter=1
+for model in $(ls -d ppcls_model/* )
+do
+    echo "[Benchmark-PPcls] ${counter}/${num_of_models} $model ..."
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend ort
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend ort
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend paddle
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend paddle
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend ov
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend ov
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend ort
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend paddle
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend trt
+    python benchmark_ppcls.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend trt --enable_trt_fp16 True
+    counter=$(($counter+1))
+    step=$(( $counter % 1 ))
+    if [ $step = 0 ]
+    then
+        wait
+    fi
+done
+
+wait
+
+rm -rf result_ppcls.txt
+touch result_ppcls.txt
+cat ppcls_model/*.txt >> ./result_ppcls.txt
+
+python convert_info.py --txt_path result_ppcls.txt --domain ppcls
diff --git a/benchmark/run_benchmark_ppdet.sh b/benchmark/run_benchmark_ppdet.sh
new file mode 100644
index 000000000..55f7de39f
--- /dev/null
+++ b/benchmark/run_benchmark_ppdet.sh
@@ -0,0 +1,33 @@
+echo "[FastDeploy]    Running PPdet benchmark..."
+
+num_of_models=$(ls -d ppdet_model/* | wc -l)
+
+counter=1
+for model in $(ls -d ppdet_model/* )
+do
+    echo "[Benchmark-PPdet] ${counter}/${num_of_models} $model ..."
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 1 --iter_num 2000 --backend ort
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 8 --iter_num 2000 --backend ort
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 1 --iter_num 2000 --backend paddle
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 8 --iter_num 2000 --backend paddle
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 1 --iter_num 2000 --backend ov
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --cpu_num_thread 8 --iter_num 2000 --backend ov
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend ort
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend paddle
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend trt
+    python benchmark_ppdet.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend trt --enable_trt_fp16 True
+    counter=$(($counter+1))
+    step=$(( $counter % 1 ))
+    if [ $step = 0 ]
+    then
+        wait
+    fi
+done
+
+wait
+
+rm -rf result_ppdet.txt
+touch result_ppdet.txt
+cat ppdet_model/*.txt >> ./result_ppdet.txt
+
+python convert_info.py --txt_path result_ppdet.txt --domain ppdet
diff --git a/benchmark/run_benchmark_ppseg.sh b/benchmark/run_benchmark_ppseg.sh
new file mode 100644
index 000000000..e878c1529
--- /dev/null
+++ b/benchmark/run_benchmark_ppseg.sh
@@ -0,0 +1,33 @@
+echo "[FastDeploy]    Running PPseg benchmark..."
+
+num_of_models=$(ls -d ppseg_model/* | wc -l)
+
+counter=1
+for model in $(ls -d ppseg_model/* )
+do
+    echo "[Benchmark-PPseg] ${counter}/${num_of_models} $model ..."
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend ort
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend ort
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend paddle
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend paddle
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 1 --iter_num 2000 --backend ov
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --cpu_num_thread 8 --iter_num 2000 --backend ov
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend ort
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend paddle
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend trt
+    python benchmark_ppseg.py --model $model --image ILSVRC2012_val_00000010.jpeg --device gpu --iter_num 2000 --backend trt --enable_trt_fp16 True
+    counter=$(($counter+1))
+    step=$(( $counter % 1 ))
+    if [ $step = 0 ]
+    then
+        wait
+    fi
+done
+
+wait
+
+rm -rf result_ppseg.txt
+touch result_ppseg.txt
+cat ppseg_model/*.txt >> ./result_ppseg.txt
+
+python convert_info.py --txt_path result_ppseg.txt --domain ppseg
diff --git a/benchmark/run_benchmark_yolo.sh b/benchmark/run_benchmark_yolo.sh
new file mode 100644
index 000000000..6f36c41eb
--- /dev/null
+++ b/benchmark/run_benchmark_yolo.sh
@@ -0,0 +1,30 @@
+echo "[FastDeploy]    Running Yolo benchmark..."
+
+num_of_models=$(ls -d yolo_model/* | wc -l)
+
+counter=1
+for model in $(ls -d yolo_model/* )
+do
+    echo "[Benchmark-Yolo] ${counter}/${num_of_models} $model ..."
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --cpu_num_thread 1 --iter_num 2000 --backend ort
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --cpu_num_thread 8 --iter_num 2000 --backend ort
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --cpu_num_thread 1 --iter_num 2000 --backend ov
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --cpu_num_thread 8 --iter_num 2000 --backend ov
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend ort
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend trt
+    python benchmark_yolo.py --model $model --image 000000014439.jpg --device gpu --iter_num 2000 --backend trt --enable_trt_fp16 True
+    counter=$(($counter+1))
+    step=$(( $counter % 1 ))
+    if [ $step = 0 ]
+    then
+        wait
+    fi
+done
+
+wait
+
+rm -rf result_yolo.txt
+touch result_yolo.txt
+cat yolo_model/*.txt >> ./result_yolo.txt
+
+python convert_info.py --txt_path result_yolo.txt --domain yolo
diff --git a/benchmark/vision/classification/ppclas/benchmark_ppclas_multi_input.py b/benchmark/vision/classification/ppclas/benchmark_ppclas_multi_input.py
deleted file mode 100644
index 3caf92982..000000000
--- a/benchmark/vision/classification/ppclas/benchmark_ppclas_multi_input.py
+++ /dev/null
@@ -1,178 +0,0 @@
-import fastdeploy as fd
-import cv2
-import os
-from tqdm import trange
-import numpy as np
-import datetime
-import json
-
-
-def parse_arguments():
-    import argparse
-    import ast
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--model", required=True, help="Path of PaddleClas model.")
-    parser.add_argument(
-        "--image", type=str, required=False, help="Path of test image file.")
-    parser.add_argument(
-        "--input_name",
-        type=str,
-        required=False,
-        default="inputs",
-        help="input name of inference file.")
-    parser.add_argument(
-        "--topk", type=int, default=1, help="Return topk results.")
-    parser.add_argument(
-        "--cpu_num_thread",
-        type=int,
-        default=12,
-        help="default number of cpu thread.")
-    parser.add_argument(
-        "--size",
-        nargs='+',
-        type=int,
-        default=[1, 3, 224, 224],
-        help="size of inference array.")
-    parser.add_argument(
-        "--iter_num",
-        required=True,
-        type=int,
-        default=30,
-        help="number of iterations for computing performace.")
-    parser.add_argument(
-        "--device",
-        nargs='+',
-        type=str,
-        default=['cpu', 'cpu', 'cpu', 'gpu', 'gpu', 'gpu'],
-        help="Type of inference device, support 'cpu' or 'gpu'.")
-    parser.add_argument(
-        "--backend",
-        nargs='+',
-        type=str,
-        default=['ort', 'paddle', 'ov', 'ort', 'trt', 'paddle'],
-        help="inference backend.")
-    args = parser.parse_args()
-    backend_list = ['ov', 'trt', 'ort', 'paddle']
-    device_list = ['cpu', 'gpu']
-    assert len(args.device) == len(
-        args.backend), "the same number of --device and --backend is requested"
-    assert args.iter_num > 10, "--iter_num has to bigger than 10"
-    assert len(args.size
-               ) == 4, "size should include 4 values, e.g., --size 1 3 300 300"
-    for b in args.backend:
-        assert b in backend_list, "%s backend is not supported" % b
-    for d in args.device:
-        assert d in device_list, "%s device is not supported" % d
-    return args
-
-
-def build_option(index, args):
-    option = fd.RuntimeOption()
-    device = args.device[index]
-    backend = args.backend[index]
-    option.set_cpu_thread_num(args.cpu_num_thread)
-    if device == "gpu":
-        option.use_gpu()
-
-    if backend == "trt":
-        assert device == "gpu", "the trt backend need device==gpu"
-        option.use_trt_backend()
-        option.set_trt_input_shape(args.input_name, args.size)
-    elif backend == "ov":
-        assert device == "cpu", "the openvino backend need device==cpu"
-        option.use_openvino_backend()
-
-    elif backend == "paddle":
-        option.use_paddle_backend()
-
-    elif backend == "ort":
-        option.use_ort_backend()
-
-    else:
-        print("%s is an unsupported backend" % backend)
-
-    print("============= inference using %s backend on %s device ============="
-          % (args.backend[index], args.device[index]))
-    return option
-
-
-args = parse_arguments()
-
-save_dict = dict()
-
-for index, device_name in enumerate(args.device):
-    if device_name not in save_dict:
-        save_dict[device_name] = dict()
-
-    # 配置runtime，加载模型
-    runtime_option = build_option(index, args)
-
-    model_file = os.path.join(args.model, "inference.pdmodel")
-    params_file = os.path.join(args.model, "inference.pdiparams")
-    config_file = os.path.join(args.model, "inference_cls.yaml")
-    model = fd.vision.classification.PaddleClasModel(
-        model_file, params_file, config_file, runtime_option=runtime_option)
-
-    # 创建要输入的向量
-    channel = args.size[1]
-    height = args.size[2]
-    width = args.size[3]
-    input_array = np.random.randint(
-        0, high=255, size=(height, width, channel), dtype=np.uint8)
-
-    # 如果有输入图片，则使用输入的图片进行推理
-    if args.image:
-        input_array = cv2.imread(args.image)
-    model_name = args.model.split('/')
-    model_name = model_name[-1] if model_name[-1] else model_name[-2]
-    print(" Model: ", model_name, " Input shape: ", input_array.shape)
-    start_time = datetime.datetime.now()
-    model.enable_record_time_of_runtime()
-    warmup_iter = args.iter_num // 5
-    warmup_end2end_time = 0
-    if "iter_num" not in save_dict:
-        save_dict["iter_num"] = args.iter_num
-    if "warmup_iter" not in save_dict:
-        save_dict["warmup_iter"] = warmup_iter
-    if "cpu_num_thread" not in save_dict:
-        save_dict["cpu_num_thread"] = args.cpu_num_thread
-    for i in trange(args.iter_num, desc="Inference Progress"):
-        if i == warmup_iter:
-            # 计算warmup端到端总时间(s)
-            warmup_time = datetime.datetime.now()
-            warmup_end2end_time = warmup_time - start_time
-            warmup_end2end_time = (
-                warmup_end2end_time.days * 24 * 60 * 60 +
-                warmup_end2end_time.seconds
-            ) * 1000 + warmup_end2end_time.microseconds / 1000
-        result = model.predict(input_array, args.topk)
-    end_time = datetime.datetime.now()
-    # 计算端到端（前处理，推理，后处理）的总时间
-    statis_info_of_runtime_dict = model.print_statis_info_of_runtime()
-    end2end_time = end_time - start_time
-    end2end_time = (end2end_time.days * 24 * 60 * 60 + end2end_time.seconds
-                    ) * 1000 + end2end_time.microseconds / 1000
-    remain_end2end_time = end2end_time - warmup_end2end_time
-    pre_post_process = end2end_time - statis_info_of_runtime_dict[
-        "total_time"] * 1000
-    end2end = remain_end2end_time / (args.iter_num - warmup_iter)
-    runtime = statis_info_of_runtime_dict["avg_time"] * 1000
-    print("Total time of end2end: %s ms" % str(end2end_time))
-    print("Average time of end2end exclude warmup step: %s ms" % str(end2end))
-    print("Total time of preprocess and postprocess in warmup step: %s ms" %
-          str(warmup_end2end_time - statis_info_of_runtime_dict["warmup_time"]
-              * 1000))
-    print(
-        "Average time of preprocess and postprocess exclude warmup step: %s ms"
-        % str((remain_end2end_time - statis_info_of_runtime_dict["remain_time"]
-               * 1000) / (args.iter_num - warmup_iter)))
-    # 结构化输出
-    backend_name = args.backend[index]
-    save_dict[device_name][backend_name] = {
-        "end2end": end2end,
-        "runtime": runtime
-    }
-    json_str = json.dumps(save_dict)
-    with open("%s.json" % model_name, 'w', encoding='utf-8') as fw:
-        json.dump(json_str, fw, indent=4, ensure_ascii=False)
diff --git a/fastdeploy/vision/detection/contrib/yolov5.cc b/fastdeploy/vision/detection/contrib/yolov5.cc
index 306051e80..0655b7f3c 100644
--- a/fastdeploy/vision/detection/contrib/yolov5.cc
+++ b/fastdeploy/vision/detection/contrib/yolov5.cc
@@ -59,8 +59,8 @@ YOLOv5::YOLOv5(const std::string& model_file, const std::string& params_file,
                const RuntimeOption& custom_option,
                const Frontend& model_format) {
   if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
-    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
+    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
diff --git a/fastdeploy/vision/detection/contrib/yolov6.cc b/fastdeploy/vision/detection/contrib/yolov6.cc
index 7c6827433..08279c482 100644
--- a/fastdeploy/vision/detection/contrib/yolov6.cc
+++ b/fastdeploy/vision/detection/contrib/yolov6.cc
@@ -63,8 +63,8 @@ YOLOv6::YOLOv6(const std::string& model_file, const std::string& params_file,
                const RuntimeOption& custom_option,
                const Frontend& model_format) {
   if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
-    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
+    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
diff --git a/fastdeploy/vision/detection/contrib/yolov7.cc b/fastdeploy/vision/detection/contrib/yolov7.cc
index edc1b9048..759b7b187 100644
--- a/fastdeploy/vision/detection/contrib/yolov7.cc
+++ b/fastdeploy/vision/detection/contrib/yolov7.cc
@@ -61,8 +61,8 @@ YOLOv7::YOLOv7(const std::string& model_file, const std::string& params_file,
                const RuntimeOption& custom_option,
                const Frontend& model_format) {
   if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
-    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
+    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
   } else {
     valid_cpu_backends = {Backend::PDINFER};
     valid_gpu_backends = {Backend::PDINFER};
diff --git a/fastdeploy/vision/detection/contrib/yolox.cc b/fastdeploy/vision/detection/contrib/yolox.cc
index 5d3880657..99542d332 100644
--- a/fastdeploy/vision/detection/contrib/yolox.cc
+++ b/fastdeploy/vision/detection/contrib/yolox.cc
@@ -75,8 +75,8 @@ void LetterBoxWithRightBottomPad(Mat* mat, std::vector<int> size,
 YOLOX::YOLOX(const std::string& model_file, const std::string& params_file,
              const RuntimeOption& custom_option, const Frontend& model_format) {
   if (model_format == Frontend::ONNX) {
-    valid_cpu_backends = {Backend::ORT};  // 指定可用的CPU后端
-    valid_gpu_backends = {Backend::ORT, Backend::TRT};  // 指定可用的GPU后端
+    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT};
+    valid_gpu_backends = {Backend::ORT, Backend::TRT};
   } else {
     valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
     valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
diff --git a/fastdeploy/vision/detection/ppdet/picodet.cc b/fastdeploy/vision/detection/ppdet/picodet.cc
index 7c961d1f8..bae51b409 100644
--- a/fastdeploy/vision/detection/ppdet/picodet.cc
+++ b/fastdeploy/vision/detection/ppdet/picodet.cc
@@ -24,7 +24,7 @@ PicoDet::PicoDet(const std::string& model_file, const std::string& params_file,
                  const RuntimeOption& custom_option,
                  const Frontend& model_format) {
   config_file_ = config_file;
-  valid_cpu_backends = {Backend::ORT, Backend::PDINFER};
+  valid_cpu_backends = {Backend::PDINFER, Backend::ORT};
   valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
diff --git a/fastdeploy/vision/detection/ppdet/ppyolo.cc b/fastdeploy/vision/detection/ppdet/ppyolo.cc
index 6c202f0d0..06b701dfb 100644
--- a/fastdeploy/vision/detection/ppdet/ppyolo.cc
+++ b/fastdeploy/vision/detection/ppdet/ppyolo.cc
@@ -23,7 +23,7 @@ PPYOLO::PPYOLO(const std::string& model_file, const std::string& params_file,
                const RuntimeOption& custom_option,
                const Frontend& model_format) {
   config_file_ = config_file;
-  valid_cpu_backends = {Backend::PDINFER};
+  valid_cpu_backends = {Backend::OPENVINO, Backend::PDINFER};
   valid_gpu_backends = {Backend::PDINFER};
   has_nms_ = true;
   runtime_option = custom_option;
diff --git a/fastdeploy/vision/detection/ppdet/yolov3.cc b/fastdeploy/vision/detection/ppdet/yolov3.cc
index 8de0ec231..35f8463ac 100644
--- a/fastdeploy/vision/detection/ppdet/yolov3.cc
+++ b/fastdeploy/vision/detection/ppdet/yolov3.cc
@@ -23,7 +23,7 @@ YOLOv3::YOLOv3(const std::string& model_file, const std::string& params_file,
                const RuntimeOption& custom_option,
                const Frontend& model_format) {
   config_file_ = config_file;
-  valid_cpu_backends = {Backend::ORT, Backend::PDINFER};
+  valid_cpu_backends = {Backend::OPENVINO, Backend::ORT, Backend::PDINFER};
   valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;
diff --git a/fastdeploy/vision/segmentation/ppseg/model.cc b/fastdeploy/vision/segmentation/ppseg/model.cc
index e63c82b7c..752dacc96 100644
--- a/fastdeploy/vision/segmentation/ppseg/model.cc
+++ b/fastdeploy/vision/segmentation/ppseg/model.cc
@@ -13,7 +13,7 @@ PaddleSegModel::PaddleSegModel(const std::string& model_file,
                                const RuntimeOption& custom_option,
                                const Frontend& model_format) {
   config_file_ = config_file;
-  valid_cpu_backends = {Backend::PDINFER};
+  valid_cpu_backends = {Backend::OPENVINO, Backend::PDINFER};
   valid_gpu_backends = {Backend::PDINFER, Backend::TRT};
   runtime_option = custom_option;
   runtime_option.model_format = model_format;