{"id":6691,"date":"2023-03-06T16:32:27","date_gmt":"2023-03-06T08:32:27","guid":{"rendered":"https:\/\/makerpro.cc\/tutors\/?post_type=docs&#038;p=6691"},"modified":"2023-06-07T23:37:46","modified_gmt":"2023-06-07T15:37:46","password":"","slug":"openvino-2022%e5%a4%a7%e6%94%b9%e7%89%88-%e8%b7%9f%e9%81%8e%e5%8e%bb%e7%9b%b8%e6%af%94%e6%9c%89%e5%93%aa%e4%ba%9b%e4%b8%8d%e4%b8%80%e6%a8%a3%e4%ba%86%e5%91%a2","status":"publish","type":"docs","link":"https:\/\/makerpro.cc\/makerlearn\/knowledgebase\/openvino-2022%e5%a4%a7%e6%94%b9%e7%89%88-%e8%b7%9f%e9%81%8e%e5%8e%bb%e7%9b%b8%e6%af%94%e6%9c%89%e5%93%aa%e4%ba%9b%e4%b8%8d%e4%b8%80%e6%a8%a3%e4%ba%86%e5%91%a2\/","title":{"rendered":"OpenVINO 2022\u5927\u6539\u7248? \u8ddf\u904e\u53bb\u76f8\u6bd4\u6709\u54ea\u4e9b\u4e0d\u4e00\u6a23\u4e86\u5462?"},"content":{"rendered":"<h3>\u9019\u6b21\u6539\u7248\u5305\u542b\u4e86\u54ea\u4e9b\u9762\u5411\u5462?<\/h3>\n<p>\u5f9e\u5b98\u65b9\u91cb\u51fa\u7684\u8aaa\u660e\u6587\u4ef6\u4e2d\u53ef\u5f97\u77e5\u9019\u6b21\u4e3b\u8981\u6539\u7248\u7684\u9762\u5411\u5305\u542b\u4ee5\u4e0b\u516d\u500b\uff1a<\/p>\n<ul>\n<li>\u63d0\u4f9b\u66f4\u591a\u6a23\u4fbf\u6377\u7684\u5b89\u88dd\u65b9\u5f0f<\/li>\n<li>\u652f\u63f4\u66f4\u591a\u985e\u578b\u786c\u9ad4\u53ca\u7570\u69cb\u63a8\u8ad6<\/li>\n<li>\u589e\u52a0\u66f4\u591a\u9810\u8a13\u7df4\u6a21\u578b\u53ca\u516c\u958b\u6a21\u578b\u5eab<\/li>\n<li>\u66f4\u7c21\u6f54\u6709\u529b\u7684API\u7a0b\u5f0f\u958b\u767c<\/li>\n<li>\u6a21\u578b\u512a\u5316\u53c3\u6578\u7c21\u5316\u53ca\u81ea\u52d5\u5316<\/li>\n<li>\u90e8\u7f72\u6642\u66f4\u667a\u80fd\u7ba1\u7406\u786c\u9ad4\u88dd\u7f6e<\/li>\n<\/ul>\n<p><img data-recalc-dims=\"1\" fetchpriority=\"high\" decoding=\"async\" class=\"aligncenter size-full wp-image-10544\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=1200%2C570&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"570\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?w=1344&amp;ssl=1 1344w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=300%2C142&amp;ssl=1 300w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=1024%2C486&amp;ssl=1 1024w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=768%2C365&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=900%2C427&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/001.png?resize=1200%2C570&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><\/p>\n<h3>\u91dd\u5c0d\u9019\u4e9b\u6539\u8b8a\u6709\u6c92\u6709\u66f4\u8a73\u7d30\u7684\u89e3\u8aaa\u5462?<\/h3>\n<p>\u4ee5\u4e0b\u6703\u6839\u64da\u4e0d\u540c\u9762\u76f8\u505a\u8aaa\u660e\uff0c<\/p>\n<p><strong>\u63d0\u4f9b\u66f4\u591a\u6a23\u4fbf\u6377\u7684\u5b89\u88dd\u65b9\u5f0f\uff1a<\/strong><\/p>\n<p>OpenVINO \u539f\u672c\u5c31\u6709\u652f\u63f4\u591a\u7a2e\u4f5c\u696d\u7cfb\u7d71\u53ca\u591a\u7a2e\u5b89\u88dd\u65b9\u5f0f\uff0c\u5728\u5b89\u88dd\u6642\u6703\u5c07\u76f8\u95dc\u5de5\u5177\u3001\u76f8\u4f9d\u5957\u4ef6\u4e00\u8d77\u5b89\u88dd\uff0c\u96d6\u7136\u9019\u6a23\u5f88\u65b9\u4fbf\u958b\u767c\uff0c\u4f46\u6703\u4f54\u7528\u592a\u591a\u5b89\u88dd\u7a7a\u9593\u3002\u6539\u7248\u5f8c\u9664\u4e86\u4fdd\u7559\u539f\u5148\u591a\u6a23\u6027\u7684\u5b89\u88dd\u5f48\u6027\u5916\uff0c\u540c\u6642\u52a0\u5165 PyPl \u4f86\u6e90\u7684\u5b89\u88dd\u65b9\u5f0f\uff0c\u8b93\u4f7f\u7528 Python \u958b\u767c\u74b0\u5883\u7684\u670b\u53cb\u80fd\u66f4\u5bb9\u6613\u5b89\u88dd OpenVINO\uff0c\u4e26\u5c07\u904b\u884c\u6240\u9700\u57fa\u790e\u5143\u4ef6 ( OpenVINO Runtime ) \u548c\u958b\u767c\u5de5\u5177\u5206\u958b\u5b89\u88dd\uff0c\u4ee5\u4fbf\u7bc0\u7701\u6642\u9593\u53ca\u7a7a\u9593\u3002<\/p>\n<div id=\"attachment_10545\" style=\"width: 1354px\" class=\"wp-caption aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" aria-describedby=\"caption-attachment-10545\" class=\"size-full wp-image-10545\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=1200%2C861&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"861\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?w=1344&amp;ssl=1 1344w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=300%2C215&amp;ssl=1 300w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=1024%2C734&amp;ssl=1 1024w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=768%2C551&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=900%2C646&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/002.png?resize=1200%2C861&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><p id=\"caption-attachment-10545\" class=\"wp-caption-text\">OpenVINO \u65b0\u820a\u7248\u672c\u5b89\u88dd\u5dee\u7570\u3002( OmniXRI \u6574\u7406\u88fd\u4f5c\uff0c2022\/7\/20)<\/p><\/div>\n<p><strong>\u652f\u63f4\u66f4\u591a\u985e\u578b\u786c\u9ad4\u53ca\u7570\u69cb\u63a8\u8ad6\uff1a<\/strong><\/p>\n<p>OpenVINO \u63a8\u51fa\u6642\u6700\u5927\u4eae\u9ede\u70ba\u5176\u53ef\u900f\u904e\u63a8\u8ad6\u5f15\u64ce\u5916\u639b ( Inference Engine Plugin ) \u4e0d\u7528\u91cd\u5beb\u7a0b\u5f0f\u5373\u53ef\u652f\u63f4\u81ea\u5bb6\u7684\u5404\u7a2e\u786c\u9ad4\u4f86\u9032\u884c\u63a8\u8ad6\uff0c\u5305\u62ec Intel Core\/Atom\/Xeon CPU\uff08AVX512\/AVX2\/SSE\u52a0\u901f\u904b\u7b97\u6307\u4ee4\u96c6\uff09\u3001iGPU\uff08Intel HD Graphics, Iris \u5167\u986f\u7e6a\u5716\u6676\u7247\uff09\u3001FPGA\uff08\u539f Altera\uff0c\u73fe\u70ba Intel Arria 10 GX\uff09\u7b49\uff0c\u9019\u5e7e\u5e74\u53c8\u9678\u7e8c\u589e\u52a0\u66f4\u591a\u7684 Plugin\uff0c\u5982 HDDL\uff08\u5373\u591a\u500b VPU\uff09\u3001Heterogeneous \u7570\u69cb\u904b\u7b97\uff08\u5373\u540c\u6642\u53ef\u4f7f\u7528\u591a\u7a2e\u786c\u9ad4\u5354\u540c\u904b\u7b97\uff09\u3002<\/p>\n<p>\u9019\u6b21\u6539\u7248\u66f4\u662f\u52a0\u5165\u5c0d Intel \u6700\u65b0\u7684\u7368\u7acb\u986f\u5361 GPU \u7684\u652f\u63f4\uff0c\u5305\u62ec Xe, Xe Pro \u7cfb\u5217\uff0c\u5927\u5927\u63d0\u5347\u63a8\u8ad6\u901f\u5ea6\u3002\u5c0d\u65bc Intel 11 \u53ca 12 \u4ee3 CPU \u800c\u8a00\uff0c\u66f4\u52a0\u5165\u5c0d AVX512-VNNI, AVX512_BF16 \u6307\u4ee4\u96c6\u7684\u652f\u63f4\uff0c\u540c\u6642\u4ea6\u6709\u652f\u63f4 CPU \u5167\u5efa\u7684 GPU Iris Xe\uff0c\u4f7f\u5f97 CPU \u7684\u904b\u7b97\u8b8a\u5f97\u66f4\u6709\u6548\u7387\u3002<\/p>\n<p><strong>\u589e\u52a0\u66f4\u591a\u9810\u8a13\u7df4\u6a21\u578b\u53ca\u516c\u958b\u6a21\u578b\u5eab\uff1a<\/strong><\/p>\n<p>OpenVINO \u70ba\u4e86\u8b93\u5927\u5bb6\u66f4\u5bb9\u6613\u4e0a\u624b\uff0c\u5f9e\u63a8\u51fa\u6642\u5c31\u6709\u63d0\u4f9b\u5f88\u591a\u7bc4\u4f8b\u7a0b\u5f0f\u3001Intel\u9810\u8a13\u7df4\u6a21\u578b\u3001\u516c\u958b\u9810\u8a13\u7df4\u6a21\u578b\u3001\u4ee5\u53ca\u958b\u6e90\u6a21\u578b\u52d5\u7269\u5712 ( Open Model Zoo ) \uff0c2021 \u6539\u7248\u6642\u5c07\u4e0a\u8ff0\u5167\u5bb9\u5f9e Resource \u5206\u9805\u4e2d\u91cd\u65b0\u6574\u5408\u5230 Model Zoo \u5206\u9805\u4e2d\uff0c\u4e26\u589e\u52a0\u4ee5 Jupyter Notebooks \u578b\u5f0f\u7684 Tutorials \u5206\u9805\u7bc4\u4f8b\u7a0b\u5f0f\uff0c\u8b93\u5927\u5bb6\u80fd\u66f4\u4e86\u89e3\u6bcf\u500b\u7a0b\u5f0f\u6bb5\u7684\u7528\u6cd5\u53ca\u5feb\u901f\u9ad4\u9a57\u904b\u884c\u63a8\u8ad6\u7d50\u679c\uff0c\u540c\u6642\u589e\u52a0\u4e86\u8a31\u591a\u5faa\u74b0\u795e\u7d93\u7db2\u8def ( Recurrent neural network, RNN ) \u76f8\u95dc\u6a21\u578b\uff0c\u5982 LSTM \u3001\u8a9e\u97f3\u8fa8\u8b58\u3001\u81ea\u7136\u8a9e\u8a00\u8655\u7406 ( Natural Language Processing, NLP ) \u7b49\u3002<\/p>\n<p>\u9019\u6b21\u6539\u7248\u5247\u5c07\u539f\u6709\u7684 Open Model Zoo Demos \u79fb\u9664 \uff08\u9700\u5206\u958b\u5b89\u88dd\uff09\uff0c\u4e26\u65b0\u589e\u4e86\u8a31\u591a\u6a21\u578b\uff0c\u5982\u4eba\u81c9\u7279\u5fb5\u3001\u82f1\u6587\u624b\u5beb\u8fa8\u8b58\u3001\u82f1\u6587\u6587\u5b57\u8fa8\u8b58\u3001\u4eba\u54e1\u5075\u6e2c\u3001\u7269\u4ef6\u5075\u6e2c\u3001\u6a5f\u5668\u7ffb\u8b6f\u3001\u80cc\u666f\u6473\u5716\u3001\u6587\u5b57\u9810\u6e2c\u3001\u5f71\u50cf\u8a9e\u7fa9\u5206\u5272\u3001\u8a9e\u97f3\u8fa8\u8b58\u7b49\u3002\u5176\u4e2d\u5305\u62ec Intel \u9810\u8a13\u7df4\u6a21\u578b 16 \u9805\u53ca\u516c\u958b\u9810\u8a13\u7df4\u6a21\u578b 15 \u9805\u3002\u53e6\u5916 Jupyter Notebook Tutorials \u4ea6\u65b0\u589e 9 \u500b\u7bc4\u4f8b\uff0c\u5305\u542b\u4eba\u9ad4\u52d5\u4f5c\u8fa8\u8b58\u3001\u8a13\u5f8c\u91cf\u5316 ( Post-Training Quantization )\u3001\u624b\u5beb\u8fa8\u8b58\u3001\u8a9e\u97f3\u8b8a\u6587\u5b57\u3001\u81ea\u7136\u8a9e\u8a00\u554f\u7b54\u7b49\u3002<\/p>\n<p><div id=\"attachment_10546\" style=\"width: 1666px\" class=\"wp-caption aligncenter\"><img data-recalc-dims=\"1\" decoding=\"async\" aria-describedby=\"caption-attachment-10546\" class=\"size-full wp-image-10546\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=1200%2C1268&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"1268\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?w=1656&amp;ssl=1 1656w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=284%2C300&amp;ssl=1 284w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=969%2C1024&amp;ssl=1 969w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=768%2C812&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=1453%2C1536&amp;ssl=1 1453w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=900%2C951&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/003.jpg?resize=1200%2C1268&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><p id=\"caption-attachment-10546\" class=\"wp-caption-text\">OpenVINO 2022.1 \u65b0\u589e\u6a21\u578b\u53ca\u6559\u7a0b\u3002( OmniXRI \u53c3\u8003 [2] \u6574\u7406\u88fd\u4f5c, 2022\/7\/20 )<\/p><\/div><strong>\u66f4\u7c21\u6f54\u6709\u529b\u7684API\u7a0b\u5f0f\u958b\u767c\uff1a<\/strong><\/p>\n<p>\u6b64\u6b21 2022.1 \u7248\u5728\u63a8\u8ad6\u6d41\u7a0b\u4e0a\u6709\u91cd\u5927\u6539\u9769\uff0c\u5c07\u539f\u5148\u7684\u7e41\u96dc\u7684\u63a8\u8ad6\u5f15\u64ce\u6539\u70ba\u55ae\u4e00\u904b\u884c\u5eab ( OpenVINO Runtime, OV )\uff0c\u4e26\u5927\u5e45\u7c21\u5316 API \u7684\u547c\u53eb\u65b9\u5f0f\uff0c\u5982\u4e0b\u5716\u4e0a\u65b9\u6240\u793a\u3002<\/p>\n<p>\u820a\u7248\u7684 OpenVINO \u63a8\u8ad6\u7a0b\u5e8f\u70ba\u521d\u59cb\u5316\u6838\u5fc3 InferenceEngine::Core \u53ca\u76f8\u95dc\u914d\u5957\u7a0b\u5f0f\u5eab\uff0c\u4f9d\u9700\u6c42\u914d\u7f6e\u6240\u9700\u7684\u8f38\u51fa\u5165\u9805\u76ee\uff0c\u63a5\u8457\u8f09\u5165\u6a21\u578b\uff0c\u5efa\u7acb\u63a8\u8ad6\u7d50\u679c\u8981\u6c42\uff0c\u5f85\u6e96\u5099\u597d\u8f38\u5165 ( GetBlob ) \u5f8c\u5c31\u53ef\u4ee5\u958b\u59cb\u9032\u884c\u63a8\u8ad6\uff0c\u6700\u5f8c\u518d\u5c07\u7d50\u679c\u8f49\u56de\u53ef\u61c9\u7528\u65bc\u5f8c\u7aef\u7a0b\u5f0f\u7684\u683c\u5f0f\uff0c\u5373\u53ef\u5b8c\u6210\u4e00\u6b21\u63a8\u8ad6\uff0c\u5982\u4e0b\u5716\u5de6\u5074\u4e0b\u65b9\u3002<\/p>\n<p>\u6b64\u6b21 2022.1 \u7248\u7684\u63a8\u8ad6\u6d41\u7a0b\u5247\u6539\u6210\u521d\u59cb\u5316\u53ea\u9700\u914d\u7f6e\u55ae\u4e00 ov::Core \u5373\u53ef\uff0c\u4e0d\u7528\u518d\u914d\u7f6e\u5176\u5b83\u3002\u63a5\u8457\u4f7f\u7528 ov::CompiledModel \u4f86\u8655\u7406\u8f09\u5165\u6a21\u578b\u53ca\u5c0d\u61c9\u63a8\u8ad6\u786c\u9ad4\uff0c\u9019\u88e1\u9810\u8a2d\u53ef\u4f7f\u7528 AUTO \u53c3\u6578\u4f86\u9032\u884c\u63a8\u8ad6\u786c\u9ad4\u81ea\u52d5\u5316\u5075\u6e2c\u53ca\u914d\u7f6e\uff0c\u63a5\u8457\u540c\u6a23\u5730\u5efa\u7acb\u63a8\u8ad6\u7d50\u679c\u8981\u6c42\uff0c\u8a2d\u7f6e\u8f38\u5165\u5167\u5bb9\u3002\u9019\u88e1\u6539\u7528 ov::Tensor \u4f86\u53d6\u4ee3\u539f\u4f86\u7684 Blob \u7528\u6cd5\uff0c\u6b64\u7a2e\u505a\u6cd5\u6703\u66f4\u63a5\u8fd1\u4e3b\u6d41\u6a21\u578b\u6846\u67b6\u7684\u63cf\u8ff0\u65b9\u5f0f\u3002<\/p>\n<div id=\"attachment_10547\" style=\"width: 1666px\" class=\"wp-caption aligncenter\"><img data-recalc-dims=\"1\" loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-10547\" class=\"size-full wp-image-10547\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=1200%2C625&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"625\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?w=1656&amp;ssl=1 1656w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=300%2C156&amp;ssl=1 300w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=1024%2C534&amp;ssl=1 1024w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=768%2C400&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=1536%2C800&amp;ssl=1 1536w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=900%2C469&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_03.jpg?resize=1200%2C625&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><p id=\"caption-attachment-10547\" class=\"wp-caption-text\">\u65b0\u820a\u7248\u672c\u57f7\u884c\u63a8\u8ad6\u6d41\u7a0b\u6bd4\u8f03\u3002( OmniXRI \u6574\u7406\u88fd\u4f5c, 2022\/7\/20 )<\/p><\/div>\n<p><strong>\u6a21\u578b\u512a\u5316\u53c3\u6578\u7c21\u5316\u53ca\u81ea\u52d5\u5316\uff1a<\/strong><\/p>\n<p>\u5728 OpenVINO \u4e2d\u6a21\u578b\u512a\u5316\u5668 ( Model Optimizer, MO ) \u4e00\u76f4\u662f\u5f88\u91cd\u8981\u7684\u4eae\u9ede\uff0c\u5b83\u53ef\u4ee5\u5e6b\u5927\u5bb6\u628a\u5404\u7a2e\u6846\u67b6\uff08\u5982 TensorFlow, PyTorch, ONNX \u2026\uff09\u7522\u751f\u7684\u6a21\u578b\u8f49\u63db\u5230\u4e2d\u4ecb\u8868\u793a ( IR ) \u6a94\uff0c\u4fbf\u65bc\u9032\u884c\u6a21\u578b\u7684\u512a\u5316\u3002<\/p>\n<p>\u6b64\u6b21\u6539\u7248\u5c0d\u65bc\u57f7\u884c\u6a21\u578b\u512a\u5316\u5668\u7684\u53c3\u6578\u4f5c\u4e86\u4e00\u4e9b\u512a\u5316\uff0c\u820a\u7248\u9808\u52a0\u5165\u8f38\u5165\u6a21\u578b\u7684\u5916\u5f62\uff0c\u800c\u65b0\u7248\u5c31\u4e0d\u9700\u8981\u4e86\uff0c\u56e0\u70ba\u904b\u884c\u6642\u6703\u81ea\u52d5\u89e3\u8b80\u6a21\u578b\u53d6\u5f97\u76f8\u95dc\u5916\u5f62\u53c3\u6578\u3002<\/p>\n<h3>\u5c0d\u65bc\u66f4\u667a\u80fd\u7ba1\u7406\u786c\u9ad4\u88dd\u7f6e\u6709\u4ec0\u9ebc\u4f8b\u5b50\u55ce?<\/h3>\n<p>\u6b64\u6b21\u6539\u7248\u5728\u667a\u80fd\u7ba1\u7406\u4e0a\u9084\u6709\u5e7e\u500b\u5c0f\u4eae\u9ede\uff0c\u8b93\u63a8\u8ad6\u6548\u80fd\u9084\u80fd\u518d\u7565\u5fae\u63d0\u5347\uff0c<\/p>\n<p><strong>1. Auto Device Plugin<\/strong><\/p>\n<p>\u820a\u7248\u4f7f\u7528 Inference Engine API \u6642\uff0c\u5fc5\u9808\u7531\u4f7f\u7528\u8005\u81ea\u5df1\u6c7a\u5b9a\u8981\u4f7f\u7528\u7684\u63a8\u8ad6\u786c\u9ad4\u3001\u914d\u7f6e\u5c0d\u61c9\u53c3\u6578\uff0c\u5c24\u5176\u5728\u591a\u88dd\u7f6e\u540c\u6642\u904b\u884c\u6642\u5c31\u66f4\u96e3\u8a2d\u5b9a\u3002\u6539\u7248\u5f8c\u5247\u4f7f\u7528 OpenVINO Runtime API\uff0c\u6703\u81ea\u52d5\u641c\u7d22\u7cfb\u7d71\u4e2d\u6240\u6709\u7b97\u529b\uff08\u5982 CPU, GPU, iGPU, VPU \u7b49\uff09\u53ca\u5176\u898f\u683c\uff0c\u4e26\u6839\u64da\u6a21\u578b\u7684\u7cbe\u5ea6\u3001\u7d50\u69cb\u53ca\u7528\u6236\u6307\u5b9a\u6027\u80fd\u8981\u6c42\u7b49\u8a0a\u606f\uff0c\u9078\u53d6\u4e26\u914d\u7f6e\u9069\u5408\u7b97\u529b\u55ae\u5143\u9032\u884c\u90e8\u7f72\u3002\u53ef\u652f\u63f4\u7684\u786c\u9ad4\u53ca\u6a21\u578b\u7cbe\u5ea6\u5982\u4e0b\u5716\u6240\u793a\u3002<\/p>\n<p><div id=\"attachment_10548\" style=\"width: 1666px\" class=\"wp-caption aligncenter\"><img data-recalc-dims=\"1\" loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-10548\" class=\"size-full wp-image-10548\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=1200%2C541&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"541\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?w=1656&amp;ssl=1 1656w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=300%2C135&amp;ssl=1 300w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=1024%2C461&amp;ssl=1 1024w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=768%2C346&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=1536%2C692&amp;ssl=1 1536w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=900%2C405&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_05.jpg?resize=1200%2C541&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><p id=\"caption-attachment-10548\" class=\"wp-caption-text\">OpenVINO 2022.1 \u81ea\u52d5\u88dd\u7f6e\u9078\u64c7\u652f\u6301\u88dd\u7f6e\u53ca\u7cbe\u5ea6\u3002( OmniXRI \u53c3\u8003 [11] \u6574\u7406\u88fd\u4f5c, 2022\/7\/20 )<\/p><\/div><strong>2. PERFORMANCE_HINT<\/strong><\/p>\n<p>\u642d\u914d\u524d\u9805\u5167\u5bb9\uff0c\u9084\u53ef\u52a0\u4e0a\u6027\u80fd\u63d0\u793a ov::hint::performance_mode \u4f86\u914d\u7f6e\u81ea\u52d5\u9078\u64c7\u88dd\u7f6e\uff0c\u4f7f\u5176\u9054\u5230\u6700\u4f73\u6027\u80fd\u3002\u76ee\u524d\u4e3b\u8981\u63d0\u4f9b\u541e\u5410\u91cf ( THROUGHPUT ) \u548c\u5ef6\u9072 ( LATENCY ) \u5169\u7a2e\u9078\u9805\u3002<\/p>\n<p>\u53e6\u5916 ov::hint::model_priority \u4ea6\u63d0\u4f9b\u624b\u52d5\u63a7\u5236 Auto-Device \u7684\u57f7\u884c\u512a\u5148\u6b0a\uff0c\u5982\u4e0b\u7bc4\u4f8b\u6240\u793a\u3002\u4e0d\u904e\u9ad8\u512a\u5148\u7d1a\u6a21\u578b\u6703\u52a0\u8f09\u5230\u9ad8\u512a\u5148\u7d1a\u88dd\u7f6e\u3002\u800c\u8f03\u4f4e\u512a\u5148\u7d1a\u7684\u6a21\u578b\u4e0d\u6703\u88ab\u52a0\u8f09\u5230\u88ab\u8f03\u9ad8\u512a\u5148\u7d1a\u7684\u6a21\u578b\u5360\u7528\u7684\u88dd\u7f6e\u4e0a\u3002<\/p>\n<p><strong>3.First Inference Latency<\/strong><\/p>\n<p>\u5982\u679c\u662f\u820a\u7248 OpenVINO \u7684\u4f7f\u7528\u8005\uff0c\u90a3\u4e00\u5b9a\u9047\u904e\u540c\u6a23\u7684\u6a21\u578b\u63a8\u8ad6\u5728 iGPU \u548c MYRIAD ( VPU ) \u904b\u884c\u6642\uff0c\u7b2c\u4e00\u6b21\u63a8\u8ad6\u6703\u5ef6\u90721~2\u5206\u9418\uff0c\u5c24\u5176\u4f7f\u7528 iGPU \u6642\u5ef6\u6642\u66f4\u70ba\u660e\u986f\uff0c\u800c\u4f7f\u7528 CPU \u5247\u5f88\u5feb\u5c31\u80fd\u904b\u884c\u8d77\u4f86\u3002\u9019\u500b\u554f\u984c\u5728\u6b64\u6b21\u6539\u7248\u5f8c\u6709\u660e\u986f\u7684\u6539\u5584\uff0c\u5982\u4e0b\u5716\u6240\u793a\uff0c\u5047\u8a2d\u6211\u5011\u8981\u4f7f\u7528 Intel GPU \u9032\u884c\u63a8\u8ad6\uff0c\u5247\u5728\u6a21\u578b\u7de8\u8b6f\u548c\u8f09\u5165\u5230 GPU \u524d\uff0c\u5148\u4f7f\u7528 CPU \u4f86\u63a8\u8ad6\uff0c\u5f85 GPU \u6e96\u5099\u597d\u4e86\u5c31\u5207\u63db\u904e\u53bb\uff0c\u4f7f\u7528\u8005\u5b8c\u5168\u7121\u611f\u3002\u5982\u6b64\u7b2c\u4e00\u6b21\u63a8\u8ad6\u5ef6\u6642\u5c31\u8b8a\u5f97\u5f88\u77ed\uff0c\u540c\u6642\u4fdd\u6709\u7b2c\u4e8c\u6b21\u4e4b\u5f8c GPU \u63a8\u8ad6\u901f\u5ea6\u7684\u63d0\u5347\u3002<\/p>\n<div id=\"attachment_10549\" style=\"width: 1664px\" class=\"wp-caption aligncenter\"><img data-recalc-dims=\"1\" loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-10549\" class=\"size-full wp-image-10549\" src=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=1200%2C686&#038;ssl=1\" alt=\"\" width=\"1200\" height=\"686\" srcset=\"https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?w=1654&amp;ssl=1 1654w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=300%2C172&amp;ssl=1 300w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=1024%2C586&amp;ssl=1 1024w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=768%2C439&amp;ssl=1 768w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=1536%2C879&amp;ssl=1 1536w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=900%2C515&amp;ssl=1 900w, https:\/\/i0.wp.com\/makerpro.cc\/makerlearn\/wp-content\/uploads\/2023\/03\/20220720_Fig_06.jpg?resize=1200%2C686&amp;ssl=1 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><p id=\"caption-attachment-10549\" class=\"wp-caption-text\">Auto-Device \u63d0\u5347\u9996\u6b21\u63a8\u8ad6\u5ef6\u9072\u793a\u610f\u5716<\/p><\/div>\n<p><strong>4.Dynamic Shapes (CPU plugin only)<\/strong><\/p>\n<p>\u5728\u6df1\u5ea6\u5b78\u7fd2\u6a21\u578b\u4e2d\u975c\u614b\u6a21\u578b\u5728\u63a8\u8ad6\u6642\u53ef\u900f\u904e\u8abf\u7528\u5851\u5f62\u51fd\u5f0f reshape() \u8abf\u6574\u8f38\u5165\u8cc7\u6599\u7684\u5f35\u91cf\u5f62\u72c0\uff0c\u4f46\u9019\u6a23\u7684\u8a08\u7b97\u662f\u975e\u5e38\u8017\u6642\u7684\u3002\u5982\u679c\u9047\u5230\u50cf\u81ea\u7136\u8a9e\u8a00\u8655\u7406\uff08\u5982 BERT\uff09\u3001\u98a8\u683c\u8f49\u79fb\u9019\u985e\u8f38\u5165\u5f35\u91cf\u7dad\u5ea6\u662f\u6703\u52d5\u614b\u8b8a\u5316\u7684\u6a21\u578b\uff0c\u8abf\u7528 reshape() \u5c31\u6703\u8b8a\u6210\u8a08\u7b97\u7684\u74f6\u9838\u3002<\/p>\n<p>\u6b64\u6b21\u6539\u7248\u958b\u59cb\u652f\u63f4\u52d5\u614b\u5851\u5f62\uff0c\u9664\u4e86\u53ef\u4ee5\u89e3\u6c7a\u8a08\u7b97\u8017\u6642\u554f\u984c\uff0c\u540c\u6642\u4e5f\u89e3\u6c7a\u4ee5\u5f80\u96e3\u4ee5\u8655\u7406\u5730\u6642\u5e8f\u578b\u5faa\u74b0\u795e\u7d93\u7db2\u8def\u6a21\u578b\u63a8\u8ad6\u554f\u984c\u3002<\/p>\n<p><strong>5.Preprocessing API<\/strong><\/p>\n<p>\u901a\u5e38\u6df1\u5ea6\u5b78\u7fd2\u7684\u6a21\u578b\u90fd\u6703\u6307\u5b9a\u8f38\u5165\u8cc7\u6599\u7684\u7dad\u5ea6\u3001\u6578\u503c\u683c\u5f0f\u3001\u5c3a\u5bf8\u5927\u5c0f\uff0c\u82e5\u8f38\u5165\u8cc7\u6599\u4e0d\u7b26\u5408\u6307\u5b9a\u683c\u5f0f\u6642\u5c31\u8981\u9032\u884c\u9810\u8655\u7406( Preprocessing )\uff0c\u820a\u7248 OpenVINO \u591a\u534a\u662f\u642d\u914d\u914d\u5957\u5b89\u88dd\u7684 OpenCV \u4f86\u5354\u52a9\u5b8c\u6210\uff0c\u96d6\u7136\u900f\u904e OpenCV \u53ef\u4ee5\u5b8c\u6210\u9810\u8655\u7406\u5de5\u4f5c\u4f46\u6548\u80fd\u4e0d\u9ad8\uff0c\u56e0\u6b64\u6539\u7248\u5f8c OpenVINO \u63d0\u4f9b\u4e86\u80fd\u4f7f\u7528 GPU\/VPU \u7b49\u66f4\u9ad8\u6548\u80fd\u7684\u786c\u9ad4\u4f86\u57f7\u884c\u9810\u8655\u7406\u53ca\u63a8\u8ad6\u5f8c\u8655\u7406\u7684\u51fd\u5f0f ov::preprocess::PrePostPocessor\u3002<\/p>\n<p>&nbsp;<\/p>\n<p>\u300b\u8cc7\u6599\u4f86\u6e90\uff1aOpenVINO 2022\u5927\u6539\u7248 \u8b93Edge AI\u73a9\u51fa\u65b0\u82b1\u6a23<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u6b21\u6539\u7248\u5305\u542b\u4e86\u54ea\u4e9b\u9762\u5411\u5462? \u5f9e\u5b98\u65b9\u91cb\u51fa\u7684\u8aaa\u660e\u6587\u4ef6\u4e2d\u53ef\u5f97\u77e5\u9019\u6b21\u4e3b\u8981\u6539\u7248\u7684\u9762\u5411\u5305\u542b\u4ee5\u4e0b\u516d\u500b\uff1a \u63d0\u4f9b\u66f4\u591a\u6a23\u4fbf\u6377\u7684\u5b89\u88dd&hellip;<\/p>\n","protected":false},"author":9827,"featured_media":0,"comment_status":"open","ping_status":"closed","template":"","meta":{"_acf_changed":false,"_bbp_topic_count":0,"_bbp_reply_count":0,"_bbp_total_topic_count":0,"_bbp_total_reply_count":0,"_bbp_voice_count":0,"_bbp_anonymous_reply_count":0,"_bbp_topic_count_hidden":0,"_bbp_reply_count_hidden":0,"_bbp_forum_subforum_count":0,"footnotes":""},"doc_category":[131,81],"doc_tag":[],"class_list":["post-6691","docs","type-docs","status-publish","hentry","doc_category-openvino","doc_category-artificial-intelligent","description-off"],"acf":[],"year_month":"2026-04","word_count":212,"total_views":0,"reactions":{"happy":0,"normal":0,"sad":0},"author_info":{"name":"MakerLearn\u8b1b\u5e2b\u7fa4","author_nicename":"makerlearn","author_url":"https:\/\/makerpro.cc\/makerlearn\/author\/makerlearn\/"},"doc_category_info":[{"term_name":"OpenVINO","term_url":"https:\/\/makerpro.cc\/makerlearn\/knowledgebase-category\/openvino\/"},{"term_name":"\u4eba\u5de5\u667a\u6167","term_url":"https:\/\/makerpro.cc\/makerlearn\/knowledgebase-category\/artificial-intelligent\/"}],"doc_tag_info":[],"jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/docs\/6691","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/docs"}],"about":[{"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/types\/docs"}],"author":[{"embeddable":true,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/users\/9827"}],"replies":[{"embeddable":true,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/comments?post=6691"}],"version-history":[{"count":1,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/docs\/6691\/revisions"}],"predecessor-version":[{"id":10550,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/docs\/6691\/revisions\/10550"}],"wp:attachment":[{"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/media?parent=6691"}],"wp:term":[{"taxonomy":"doc_category","embeddable":true,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/doc_category?post=6691"},{"taxonomy":"doc_tag","embeddable":true,"href":"https:\/\/makerpro.cc\/makerlearn\/wp-json\/wp\/v2\/doc_tag?post=6691"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}