{"id":58426,"date":"2025-08-16T10:15:35","date_gmt":"2025-08-16T02:15:35","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/58426.html"},"modified":"2025-08-16T10:15:35","modified_gmt":"2025-08-16T02:15:35","slug":"%e3%80%90%e9%99%84%e6%ba%90%e7%a0%81%e3%80%91yolov8-%e4%b8%80%e6%9d%a1%e9%be%99%e5%ae%9e%e6%88%98%ef%bc%9apytorch-%e2%86%92-onnx-%e2%86%92-tensorrt%ef%bc%88%e8%be%93%e5%87%ba-1x84x8400","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/58426.html","title":{"rendered":"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09"},"content":{"rendered":"

\u76ee\u5f55<\/p>\n

NVIDIA YOLOv8 \u6a21\u578b\u8f6c\u6362\u5168\u6d41\u7a0b（PyTorch \u2192 ONNX \u2192 TensorRT）<\/p>\n

\u4e00\u3001 \u73af\u5883\u51c6\u5907<\/p>\n

\u4e8c\u3001\u4ece PyTorch \u5bfc\u51fa ONNX \u6a21\u578b<\/p>\n

\u4e09\u3001\u00a0\u4f7f\u7528 ONNX Runtime \u6d4b\u8bd5\u6a21\u578b<\/p>\n

\u56db\u3001\u5c06 ONNX \u8f6c\u6362\u4e3a TensorRT<\/p>\n

4.1\u00a0trtexec \u6781\u901f\u7248<\/p>\n

4.2\u00a0 Python API（\u63a8\u8350，\u53ef\u81ea\u5b9a\u4e49\u540e\u5904\u7406）<\/p>\n

\u4e94\u3001TensorRT \u63a8\u7406\u6d4b\u8bd5（\u8f93\u51fa 1\u00d784\u00d78400 \u89e3\u7801）<\/p>\n

\u516d\u3001\u8fd0\u884c\u6548\u679c\u4e0e\u901f\u5ea6\u5bf9\u6bd4<\/p>\n

📌 \u603b\u7ed3<\/p>\n

\n

NVIDIA YOLOv8 \u6a21\u578b\u8f6c\u6362\u5168\u6d41\u7a0b（PyTorch \u2192 ONNX \u2192 TensorRT）<\/h2>\n

📢 \u672c\u6587\u8be6\u7ec6\u8bb0\u5f55\u4e86 YOLOv8 \u4ece PyTorch \u8bad\u7ec3\u6a21\u578b\u5230 ONNX，\u518d\u5230 TensorRT \u52a0\u901f\u5f15\u64ce\u7684\u5b8c\u6574\u8f6c\u6362\u8fc7\u7a0b，\u5e76\u89e3\u51b3\u4e86 ONNX \u8f93\u51fa\u5f62\u72b6\u4e3a 1\u00d784\u00d78400 \u7684\u60c5\u51b5\u3002 \u73af\u5883\u57fa\u4e8e NVIDIA GPU + PyTorch + TensorRT，\u6587\u672b\u9644\u5b8c\u6574 Python \u6e90\u7801\u3002<\/p>\n

\n

\u4e00\u3001 \u73af\u5883\u51c6\u5907<\/h3>\n

\u5728\u5f00\u59cb\u4e4b\u524d，\u5148\u786e\u4fdd\u4ee5\u4e0b\u73af\u5883\u5df2\u6b63\u786e\u5b89\u88c5：<\/p>\n\n \u73af\u5883<\/p>\n\n\u7ec4\u4ef6\u7248\u672c\u5907\u6ce8<\/tr>\n\n\n\n\n\n\n\n
Ubuntu \/ Win11<\/td>\n22.04\u00a0<\/td>\n\u5747\u53ef<\/td>\n<\/tr>\n
CUDA<\/td>\n11.8<\/td>\n\u5411\u4e0b\u517c\u5bb9\u00a0<\/td>\n<\/tr>\n
cuDNN<\/td>\n8.9.0<\/td>\ncuDNN Archive | NVIDIA Developer<\/td>\n<\/tr>\n
TensorRT<\/td>\n8.6.1<\/td>\n\u5b98\u65b9\u4e0b\u8f7d<\/td>\n<\/tr>\n
PyTorch<\/td>\n2.1.0+cu121<\/td>\n\u5efa\u8bae conda<\/td>\n<\/tr>\n
ultralytics<\/td>\n8.0.73<\/td>\n\u5b98\u65b9\u4ed3\u5e93<\/td>\n<\/tr>\n
onnx \/ onnxsim<\/td>\n1.15 \/ 0.4<\/td>\n\u5bfc\u51fa\u540e\u7528 onnxsim \u7cbe\u7b80<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

💡 \u6ce8\u610f： TensorRT \u5b89\u88c5\u5305\u9700\u8981\u548c CUDA \u7248\u672c\u5339\u914d，\u5426\u5219\u4f1a\u62a5\u9519\u3002<\/p>\n

\u5982\u679c\u4e0d\u4f1a\u5b89\u88c5，\u8bf7\u53c2\u8003\u535a\u4e3b\u7684\u53e6\u5916\u51e0\u7bc7\u535a\u6587\u3002\u3010\u4fdd\u59c6\u7ea7\u3011Windows \u7cfb\u7edf\u5b89\u88c5 TensorRT 8.6.1 \u5b89\u88c5\u6307\u5357-CSDN\u535a\u5ba2<\/p>\n

\u3010\u4fdd\u59c6\u7ea7\u3011\u5728Windows\u7cfb\u7edf\u73af\u5883\u4e0b\u5b89\u88c5CUDA 11.8\u548ccuDNN v8.9.0\u7684\u8be6\u7ec6\u6307\u5357，\u914d\u5957tensorrt8.6\u4f7f\u7528-CSDN\u535a\u5ba2<\/p>\n

\n

\u4e8c\u3001\u4ece PyTorch \u5bfc\u51fa ONNX \u6a21\u578b<\/h3>\n

YOLOv8 \u5b98\u65b9\u5df2\u7ecf\u63d0\u4f9b\u4e86\u65b9\u4fbf\u7684 export \u63a5\u53e3，\u8fd9\u91cc\u76f4\u63a5\u8c03\u7528\u5373\u53ef\u3002<\/p>\n

from ultralytics import YOLO<\/p>\n

# 1. \u52a0\u8f7d PyTorch \u683c\u5f0f\u7684\u6743\u91cd
\nmodel = YOLO("yolov8n.pt") # \u8fd9\u91cc\u53ef\u4ee5\u66ff\u6362\u4e3a\u81ea\u5df1\u7684 .pt \u6587\u4ef6<\/p>\n

# 2. \u5bfc\u51fa\u4e3a ONNX
\nmodel.export(format="onnx", opset=12, dynamic=False, simplify=True, imgsz=640)<\/p>\n

print("ONNX \u6a21\u578b\u5bfc\u51fa\u5b8c\u6210！")<\/p>\n

\u6267\u884c\u5b8c\u6210\u540e，\u4f1a\u751f\u6210 yolov8n.onnx \u6587\u4ef6，\u8f93\u51fa\u5f62\u72b6\u662f：<\/p>\n

(1, 84, 8400)<\/p>\n

\"\"：<\/p>\n

    \n
  • \n

    1 \u2192 batch size<\/p>\n<\/li>\n

  • \n

    84 \u2192 \u6bcf\u4e2a\u76ee\u6807\u7684\u7c7b\u522b\u6570 + 4 \u4e2a bbox \u5750\u6807 (COCO \u4e3a 80 \u7c7b，\u6240\u4ee5 4 + 80 = 84)<\/p>\n<\/li>\n

  • \n

    8400 \u2192 3 \u4e2a\u68c0\u6d4b\u5c42\u603b anchor \u6570（80\u00d780 + 40\u00d740 + 20\u00d720 \u4e58\u4ee5 3）<\/p>\n<\/li>\n<\/ul>\n

    \n

    \u4e09\u3001\u00a0\u4f7f\u7528 ONNX Runtime \u6d4b\u8bd5\u6a21\u578b<\/h3>\n

    \u5bfc\u51fa\u540e\u6211\u4eec\u5148\u7528 onnxruntime \u6d4b\u8bd5\u4e00\u4e0b，\u786e\u4fdd\u6a21\u578b\u80fd\u6b63\u5e38\u63a8\u7406\u3002<\/p>\n

    import onnxruntime as ort
    \nimport numpy as np<\/p>\n

    # 1. \u521b\u5efa\u63a8\u7406 Session
    \nort_session = ort.InferenceSession("yolov8n.onnx", providers=['CUDAExecutionProvider'])<\/p>\n

    # 2. \u6784\u9020\u6d4b\u8bd5\u8f93\u5165
    \ndummy_input = np.random.randn(1, 3, 640, 640).astype(np.float32)<\/p>\n

    # 3. \u63a8\u7406
    \noutputs = ort_session.run(None, {ort_session.get_inputs()[0].name: dummy_input})<\/p>\n

    print("ONNX \u8f93\u51fa shape:", outputs[0].shape) # \u5e94\u8be5\u662f (1, 84, 8400)<\/p>\n

    \"\"<\/p>\n

    \u5982\u679c\u5c0f\u4f19\u4f34\u60f3\u76f4\u63a5\u7528onnx\u63a8\u7406，\u90a3\u53ef\u4ee5\u53c2\u8003\u6211\u7684\u8fd9\u4e00\u7bc7\u535a\u5ba2\u3002\u3010\u4fdd\u59c6\u7ea7\u6559\u7a0b\u3011YOLOv8 PyTorch \u6a21\u578b\u8f6c ONNX \u5e76\u63a8\u7406，\u5305\u542b\u56fe\u50cfpreprocess\u548cnms\u7684postprocess\u5168\u6d41\u7a0b\u8d85\u8be6\u7ec6！（\u5c0f\u767d\u4e5f\u80fd\u5b66\u4f1a）-CSDN\u535a\u5ba2<\/p>\n

    \n

    \u56db\u3001\u5c06 ONNX \u8f6c\u6362\u4e3a TensorRT<\/h3>\n

    4.1\u00a0trtexec \u6781\u901f\u7248<\/h4>\n

    # FP16
    \ntrtexec –onnx=yolov8n.onnx \\\\
    \n –saveEngine=yolov8n_fp16.trt \\\\
    \n –fp16 \\\\
    \n –workspace=4096<\/p>\n

    # INT8（\u9700\u8981\u6821\u51c6\u6570\u636e，\u89c1\u4e0b\u8282）
    \ntrtexec –onnx=yolov8n.onnx \\\\
    \n –saveEngine=yolov8n_int8.trt \\\\
    \n –int8 \\\\
    \n –calib=calib \\\\
    \n –workspace=4096 <\/p>\n

    4.2\u00a0 Python API（\u63a8\u8350，\u53ef\u81ea\u5b9a\u4e49\u540e\u5904\u7406）<\/h4>\n

    \u6211\u4eec\u7528 TensorRT Python API \u5b8c\u6210\u8f6c\u6362\u3002<\/p>\n

    import tensorrt as trt
    \n# import pycuda.driver as cuda
    \n# import pycuda.autoinit
    \n# import numpy as np
    \n# import os, glob, cv2<\/p>\n

    TRT_LOGGER = trt.Logger(trt.Logger.INFO)<\/p>\n

    EXPLICIT_BATCH = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)<\/p>\n

    def GiB(x):
    \n return x * 1 << 30<\/p>\n

    def build_engine(onnx_file_path, engine_file_path, mode='FP16'):
    \n builder = trt.Builder(TRT_LOGGER)
    \n network = builder.create_network(EXPLICIT_BATCH)
    \n config = builder.create_builder_config()
    \n parser = trt.OnnxParser(network, TRT_LOGGER)<\/p>\n

    config.max_workspace_size = GiB(4)<\/p>\n

    if mode == 'FP16':
    \n config.set_flag(trt.BuilderFlag.FP16)
    \n elif mode == 'INT8':
    \n config.set_flag(trt.BuilderFlag.INT8)
    \n # \u8fd9\u91cc\u53ef\u4ee5\u7ed1\u5b9a\u81ea\u5b9a\u4e49 calibrator<\/p>\n

    with open(onnx_file_path, 'rb') as model:
    \n if not parser.parse(model.read()):
    \n for error in range(parser.num_errors):
    \n print(parser.get_error(error))
    \n return None<\/p>\n

    engine = builder.build_engine(network, config)
    \n with open(engine_file_path, 'wb') as f:
    \n f.write(engine.serialize())
    \n print(f"Build {mode} engine done!")<\/p>\n

    if __name__ == '__main__':
    \n build_engine('yolov8n.onnx', 'yolov8n_fp16.engine', mode='FP16') <\/p>\n

    \"\"<\/p>\n

    \n

    \u4e94\u3001TensorRT \u63a8\u7406\u6d4b\u8bd5（\u8f93\u51fa 1\u00d784\u00d78400 \u89e3\u7801）<\/h3>\n

    TensorRT Engine \u62ff\u5230\u7684\u662f raw tensor，\u9700\u8981\u624b\u52a8\u540e\u5904\u7406\u3002<\/p>\n

    import tensorrt as trt
    \nimport pycuda.driver as cuda
    \nimport pycuda.autoinit
    \nimport numpy as np
    \nimport cv2
    \n# import torch
    \nclass YOLOv8TRT:
    \n def __init__(self, engine_path):
    \n logger = trt.Logger(trt.Logger.INFO)
    \n with open(engine_path, 'rb') as f, trt.Runtime(logger) as runtime:
    \n self.engine = runtime.deserialize_cuda_engine(f.read())
    \n self.context = self.engine.create_execution_context()
    \n self.stream = cuda.Stream()<\/p>\n

    # \u7ed1\u5b9a\u8f93\u5165\u8f93\u51fa
    \n self.bindings = []
    \n for binding in self.engine:
    \n size = trt.volume(self.engine.get_binding_shape(binding)) * 1
    \n dtype = trt.nptype(self.engine.get_binding_dtype(binding))
    \n host_mem = cuda.pagelocked_empty(size, dtype)
    \n device_mem = cuda.mem_alloc(host_mem.nbytes)
    \n self.bindings.append(int(device_mem))
    \n if self.engine.binding_is_input(binding):
    \n self.host_input = host_mem
    \n self.device_input = device_mem
    \n else:
    \n self.host_output = host_mem
    \n self.device_output = device_mem<\/p>\n

    def infer(self, img_path):
    \n img = cv2.imread(img_path)
    \n img_in = cv2.resize(img, (640, 640))
    \n img_in = img_in[:, :, ::-1].transpose(2,0,1).astype(np.float32) \/ 255.0
    \n img_in = np.ascontiguousarray(img_in[None])
    \n np.copyto(self.host_input, img_in.ravel())<\/p>\n

    cuda.memcpy_htod_async(self.device_input, self.host_input, self.stream)
    \n self.context.execute_async_v2(self.bindings, self.stream.handle)
    \n cuda.memcpy_dtoh_async(self.host_output, self.device_output, self.stream)
    \n self.stream.synchronize()<\/p>\n

    preds = self.host_output.reshape(1, 84, 8400)
    \n return preds, img<\/p>\n

    # \u540e\u5904\u7406：\u7f6e\u4fe1\u5ea6\u8fc7\u6ee4 + NMS
    \ndef postprocess(preds, conf_thres=0.25, iou_thres=0.45):
    \n pass<\/p>\n

    if __name__ == '__main__':
    \n yolo = YOLOv8TRT('yolov8n_fp16.engine')
    \n preds, img = yolo.infer('bus.jpg')
    \n print(preds.shape)
    \n #\u540e\u7eed\u9700\u8981\u505aNMS\u540e\u5904\u7406 <\/p>\n

    \"\"<\/p>\n

    \u5982\u679c\u4f60\u4e0d\u77e5\u9053\u540e\u7eedpostprocess\u600e\u4e48\u5199，\u53ef\u4ee5\u53c2\u8003\u535a\u4e3b\u4e0a\u7bc7\u63a8\u7406onnx\u7684\u535a\u5ba2\u3002\u3010\u4fdd\u59c6\u7ea7\u6559\u7a0b\u3011YOLOv8 PyTorch \u6a21\u578b\u8f6c ONNX \u5e76\u63a8\u7406，\u5305\u542b\u56fe\u50cfpreprocess\u548cnms\u7684postprocess\u5168\u6d41\u7a0b\u8d85\u8be6\u7ec6！（\u5c0f\u767d\u4e5f\u80fd\u5b66\u4f1a）-CSDN\u535a\u5ba2<\/p>\n

    \u5982\u679c\u8fd8\u662f\u4e0d\u4f1a，\u5173\u6ce8\u7559\u8a00，\u535a\u4e3b\u624b\u628a\u624b\u6559\u5b66\u3002<\/p>\n

    \u5982\u679c\u53d1\u73b0pycuda\u5b89\u88c5\u5931\u8d25，\u51fa\u73b0\"\"\u5b89\u88c5\u4e00\u4e0b，\u91cd\u65b0\u542f\u52a8\u7535\u8111，\u518d\u6b21\u6697\u8f6c\u5373\u53ef\u3002\"\"<\/p>\n

    \n

    \u516d\u3001\u8fd0\u884c\u6548\u679c\u4e0e\u901f\u5ea6\u5bf9\u6bd4<\/h3>\n\n\u6a21\u578b\u683c\u5f0f\u63a8\u7406\u65f6\u95f4 (640×640, batch=1)\u52a0\u901f\u500d\u7387<\/tr>\n\n\n\n\n
    PyTorch<\/td>\n12.5 ms<\/td>\n1.0\u00d7<\/td>\n<\/tr>\n
    ONNX<\/td>\n10.8 ms<\/td>\n1.16\u00d7<\/td>\n<\/tr>\n
    TensorRT FP16<\/td>\n4.3 ms<\/td>\n2.9\u00d7<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u26a1 \u5728 NVIDIA RTX 3060 \u4e0a，TensorRT FP16 \u63a8\u7406\u901f\u5ea6\u63d0\u5347\u63a5\u8fd1 3 \u500d，\u663e\u5b58\u5360\u7528\u4e5f\u6709\u6240\u964d\u4f4e\u3002<\/p>\n

    \n

    📌 \u603b\u7ed3<\/h3>\n

    \u672c\u6587\u5b9e\u73b0\u4e86 YOLOv8 \u4ece PyTorch \u2192 ONNX \u2192 TensorRT \u7684\u5b8c\u6574\u6d41\u7a0b，\u652f\u6301\u8f93\u51fa 1\u00d784\u00d78400，\u5e76\u5728 GPU \u4e0a\u5b9e\u73b0\u4e86\u663e\u8457\u52a0\u901f\u3002 \u5982\u679c\u4f60\u9700\u8981\u5728 Jetson\u3001\u8fb9\u7f18\u8ba1\u7b97\u8bbe\u5907 \u4e0a\u90e8\u7f72 YOLOv8，\u8fd9\u5957\u65b9\u6cd5\u540c\u6837\u9002\u7528，\u53ea\u9700\u66f4\u6362 TensorRT \u7684\u5b89\u88c5\u7248\u672c\u5373\u53ef\u3002<\/p>\n

    \n","protected":false},"excerpt":{"rendered":"

    \u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb431\u6b21\uff0c\u70b9\u8d5e8\u6b21\uff0c\u6536\u85cf4\u6b21\u3002\u672c\u6587\u8be6\u7ec6\u4ecb\u7ecd\u4e86YOLOv8\u6a21\u578b\u4ecePyTorch\u5230ONNX\u518d\u5230TensorRT\u7684\u5b8c\u6574\u8f6c\u6362\u6d41\u7a0b\u3002\u5185\u5bb9\u6db5\u76d6\u73af\u5883\u914d\u7f6e\uff08Ubuntu\/Win11+CUDA11.8+TensorRT8.6.1\uff09\u3001\u6a21\u578b\u5bfc\u51fa\uff08PyTorch\u2192ONNX\uff09\u3001ONNX\u63a8\u7406\u6d4b\u8bd5\u3001TensorRT\u8f6c\u6362\uff08\u5305\u542bFP16\/INT8\u6a21\u5f0f\uff09\u4ee5\u53ca\u6700\u7ec8\u63a8\u7406\u5b9e\u73b0\u3002\u91cd\u70b9\u89e3\u51b3\u4e86ONNX\u8f93\u51fa\u5f62\u72b6\u4e3a1\u00d784\u00d78400\u7684\u60c5\u51b5\uff0c\u5e76\u63d0\u4f9b\u4e86Python API\u8f6c\u6362\u65b9\u6cd5\u3002<\/p>\n","protected":false},"author":2,"featured_media":58420,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[152,156,6159,50],"topic":[],"class_list":{"0":"post-58426","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","6":"hentry","7":"category-server","8":"tag-pytorch","9":"tag-yolo","11":"tag-50"},"yoast_head":"\n\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/58426.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb431\u6b21\uff0c\u70b9\u8d5e8\u6b21\uff0c\u6536\u85cf4\u6b21\u3002\u672c\u6587\u8be6\u7ec6\u4ecb\u7ecd\u4e86YOLOv8\u6a21\u578b\u4ecePyTorch\u5230ONNX\u518d\u5230TensorRT\u7684\u5b8c\u6574\u8f6c\u6362\u6d41\u7a0b\u3002\u5185\u5bb9\u6db5\u76d6\u73af\u5883\u914d\u7f6e\uff08Ubuntu\/Win11+CUDA11.8+TensorRT8.6.1\uff09\u3001\u6a21\u578b\u5bfc\u51fa\uff08PyTorch\u2192ONNX\uff09\u3001ONNX\u63a8\u7406\u6d4b\u8bd5\u3001TensorRT\u8f6c\u6362\uff08\u5305\u542bFP16\/INT8\u6a21\u5f0f\uff09\u4ee5\u53ca\u6700\u7ec8\u63a8\u7406\u5b9e\u73b0\u3002\u91cd\u70b9\u89e3\u51b3\u4e86ONNX\u8f93\u51fa\u5f62\u72b6\u4e3a1\u00d784\u00d78400\u7684\u60c5\u51b5\uff0c\u5e76\u63d0\u4f9b\u4e86Python API\u8f6c\u6362\u65b9\u6cd5\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/58426.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-08-16T02:15:35+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/08\/20250816021522-689fe9ba1af8f.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/58426.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/58426.html\",\"name\":\"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-08-16T02:15:35+00:00\",\"dateModified\":\"2025-08-16T02:15:35+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/58426.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/58426.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/58426.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/58426.html","og_locale":"zh_CN","og_type":"article","og_title":"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb431\u6b21\uff0c\u70b9\u8d5e8\u6b21\uff0c\u6536\u85cf4\u6b21\u3002\u672c\u6587\u8be6\u7ec6\u4ecb\u7ecd\u4e86YOLOv8\u6a21\u578b\u4ecePyTorch\u5230ONNX\u518d\u5230TensorRT\u7684\u5b8c\u6574\u8f6c\u6362\u6d41\u7a0b\u3002\u5185\u5bb9\u6db5\u76d6\u73af\u5883\u914d\u7f6e\uff08Ubuntu\/Win11+CUDA11.8+TensorRT8.6.1\uff09\u3001\u6a21\u578b\u5bfc\u51fa\uff08PyTorch\u2192ONNX\uff09\u3001ONNX\u63a8\u7406\u6d4b\u8bd5\u3001TensorRT\u8f6c\u6362\uff08\u5305\u542bFP16\/INT8\u6a21\u5f0f\uff09\u4ee5\u53ca\u6700\u7ec8\u63a8\u7406\u5b9e\u73b0\u3002\u91cd\u70b9\u89e3\u51b3\u4e86ONNX\u8f93\u51fa\u5f62\u72b6\u4e3a1\u00d784\u00d78400\u7684\u60c5\u51b5\uff0c\u5e76\u63d0\u4f9b\u4e86Python API\u8f6c\u6362\u65b9\u6cd5\u3002","og_url":"https:\/\/www.wsisp.com\/helps\/58426.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-08-16T02:15:35+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/08\/20250816021522-689fe9ba1af8f.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/58426.html","url":"https:\/\/www.wsisp.com\/helps\/58426.html","name":"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-08-16T02:15:35+00:00","dateModified":"2025-08-16T02:15:35+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/58426.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/58426.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/58426.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u3010\u9644\u6e90\u7801\u3011YOLOv8 \u4e00\u6761\u9f99\u5b9e\u6218\uff1aPyTorch \u2192 ONNX \u2192 TensorRT\uff08\u8f93\u51fa 1\u00d784\u00d78400 \u5168\u6d41\u7a0b\u8e29\u5751\u8bb0\u5f55\uff09"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/58426","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=58426"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/58426\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/58420"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=58426"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=58426"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=58426"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=58426"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}