{"id":1069830,"date":"2022-07-26T22:49:00","date_gmt":"2022-07-27T05:49:00","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&#038;p=1069830"},"modified":"2024-09-25T01:34:33","modified_gmt":"2024-09-25T08:34:33","slug":"layoutlmv3","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/layoutlmv3\/","title":{"rendered":"\u6587\u6863\u667a\u80fd\u591a\u6a21\u6001\u9884\u8bad\u7ec3\u6a21\u578bLayoutLMv3\uff1a\u517c\u5177\u901a\u7528\u6027\u4e0e\u4f18\u8d8a\u6027"},"content":{"rendered":"\n<p>\u7f16\u8005\u6309\uff1a\u4f01\u4e1a\u6570\u5b57\u5316\u8f6c\u578b\u4e2d\uff0c\u4ee5\u6587\u6863\u3001\u56fe\u50cf\u7b49\u591a\u6a21\u6001\u5f62\u5f0f\u4e3a\u8f7d\u4f53\u7684\u7ed3\u6784\u5316\u5206\u6790\u548c\u5185\u5bb9\u63d0\u53d6\u662f\u5176\u4e2d\u7684\u5173\u952e\u4e00\u73af\uff0c\u5feb\u901f\u3001\u81ea\u52a8\u3001\u7cbe\u51c6\u5730\u5904\u7406\u5305\u62ec\u5408\u540c\u3001\u7968\u636e\u3001\u62a5\u544a\u7b49\u4fe1\u606f\uff0c\u5bf9\u63d0\u5347\u73b0\u4ee3\u4f01\u4e1a\u751f\u4ea7\u6548\u7387\u81f3\u5173\u91cd\u8981\u3002\u56e0\u6b64\uff0c\u6587\u6863\u667a\u80fd\u6280\u672f\u5e94\u8fd0\u800c\u751f\u3002\u8fc7\u53bb\u51e0\u5e74\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u63a8\u51fa\u4e86\u901a\u7528\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3 LayoutLM \u7cfb\u5217\u7814\u7a76\u6210\u679c\uff0c\u5e76\u4e0d\u65ad\u4f18\u5316\u6a21\u578b\u5bf9\u6587\u6863\u4e2d\u6587\u672c\u3001\u5e03\u5c40\u548c\u89c6\u89c9\u4fe1\u606f\u7684\u9884\u8bad\u7ec3\u6027\u80fd\u3002\u8fd1\u671f\u53d1\u8868\u7684\u6700\u65b0\u7684 LayoutLM 3.0 \u7248\u672c\uff0c\u5728\u4ee5\u6587\u672c\u548c\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e0a\u6709\u4e86\u66f4\u52a0\u51fa\u8272\u7684\u8868\u73b0\uff0c\u8ba9\u6587\u6863\u7406\u89e3\u6a21\u578b\u5411\u8de8\u6a21\u6001\u5bf9\u9f50\u8fc8\u51fa\u4e00\u5927\u6b65\uff01<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u968f\u7740\u5404\u884c\u5404\u4e1a\u7684\u6570\u5b57\u5316\u8f6c\u578b\uff0c\u6db5\u76d6\u8868\u5355\u3001\u7968\u636e\u3001\u90ae\u4ef6\u3001\u5408\u540c\u3001\u62a5\u544a\u3001\u8bba\u6587\u7b49\u7684\u7535\u5b50\u6587\u6863\u6570\u91cf\u6301\u7eed\u589e\u957f\u3002\u7535\u5b50\u6587\u6863\u5305\u542b\u5927\u91cf\u4e0e\u884c\u4e1a\u76f8\u5173\u7684\u56fe\u50cf\u548c\u6587\u672c\u4fe1\u606f\uff0c\u4eba\u5de5\u5904\u7406\u8fd9\u4e9b\u5927\u91cf\u7684\u4fe1\u606f\u8017\u65f6\u957f\u3001\u6210\u672c\u5927\u3002\u7535\u5b50\u6587\u6863\u7684\u81ea\u52a8\u8bc6\u522b\u3001\u7406\u89e3\u548c\u5206\u6790\u6280\u672f\u5bf9\u63d0\u9ad8\u4e2a\u4eba\u6216\u4f01\u4e1a\u7684\u751f\u4ea7\u529b\u5341\u5206\u91cd\u8981\uff0c\u56e0\u6b64\u6587\u6863\u667a\u80fd\u6280\u672f\u5e94\u8fd0\u800c\u751f\u3002\u6587\u6863\u667a\u80fd\u5229\u7528\u8ba1\u7b97\u673a\u81ea\u52a8\u8bc6\u522b\u3001\u7406\u89e3\u53ca\u5206\u6790\u7535\u5b50\u6587\u6863\uff0c\u5927\u5927\u63d0\u5347\u4e86\u4e2a\u4eba\u548c\u4f01\u4e1a\u5904\u7406\u7535\u5b50\u6587\u6863\u7684\u751f\u4ea7\u529b\uff0c\u662f\u81ea\u7136\u8bed\u8a00\u5904\u7406\u548c\u8ba1\u7b97\u673a\u89c6\u89c9\u4ea4\u53c9\u9886\u57df\u7684\u4e00\u4e2a\u91cd\u8981\u7814\u7a76\u65b9\u5411\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"655\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-1024x655.png\" alt=\"\u56fe1\uff1a\u6587\u6863\u667a\u80fd\u4efb\u52a1\u793a\u4f8b\uff1a\u8868\u5355\u7406\u89e3\u3001\u6587\u6863\u5e03\u5c40\u5206\u6790\" class=\"wp-image-1070829\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-1024x655.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-300x192.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-768x491.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-1536x983.png 1536w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-2048x1310.png 2048w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-1-240x154.png 240w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe1\uff1a\u6587\u6863\u667a\u80fd\u4efb\u52a1\u793a\u4f8b\uff1a\u8868\u5355\u7406\u89e3\u3001\u6587\u6863\u5e03\u5c40\u5206\u6790<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u867d\u7136\u4e3a\u7279\u5b9a\u4efb\u52a1\u8bbe\u8ba1\u7684\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u9488\u5bf9\u67d0\u4e00\u9879\u6587\u6863\u7406\u89e3\u4efb\u52a1\u80fd\u53d6\u5f97\u8f83\u597d\u7684\u6027\u80fd\uff0c\u4f46\u8fd9\u4e9b\u65b9\u6cd5\u901a\u5e38\u4f9d\u8d56\u4e8e\u6709\u9650\u7684\u6807\u6ce8\u6570\u636e\uff0c\u800c\u5bf9\u4e8e\u6587\u6863\u7406\u89e3\u4efb\u52a1\u5c24\u5176\u662f\u5176\u4e2d\u7684\u4fe1\u606f\u62bd\u53d6\u4efb\u52a1\u6765\u8bf4\uff0c\u83b7\u53d6\u8be6\u7ec6\u7684\u6807\u6ce8\u6570\u636e\u6602\u8d35\u4e14\u8017\u65f6\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u4eba\u5458\u5c06\u76ee\u5149\u8f6c\u5411\u4e86\u90a3\u4e9b\u88ab\u5ffd\u89c6\u7684\u65e0\u6807\u6ce8\u6570\u636e\u4e0a\uff0c\u901a\u8fc7\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u6280\u672f\u6765\u5229\u7528\u73b0\u5b9e\u751f\u6d3b\u4e2d\u5927\u91cf\u7684\u65e0\u6807\u6ce8\u6570\u636e\u3002\u7531\u4e8e\u8fd1\u5e74\u6765\u9884\u8bad\u7ec3\u5728\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u88ab\u5e7f\u6cdb\u5e94\u7528\uff0c\u6240\u4ee5\u8be5\u6280\u672f\u5728\u6587\u6863\u667a\u80fd\u9886\u57df\u4e5f\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\u3002\u7ecf\u8fc7\u9884\u8bad\u7ec3\u7684\u6587\u6863\u667a\u80fd\u6a21\u578b\u53ef\u4ee5\u89e3\u6790\u5e76\u63d0\u53d6\u6587\u6863\u7684\u5404\u7c7b\u4fe1\u606f\uff0c\u8fd9\u5bf9\u6587\u6863\u667a\u80fd\u7684\u5b66\u672f\u7814\u7a76\u548c\u751f\u4ea7\u5e94\u7528\u90fd\u6709\u91cd\u8981\u610f\u4e49\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"layoutlmv3-\u8ddd\u79bb\u6587\u6863\u7406\u89e3\u8de8\u6a21\u6001\u5bf9\u9f50\u53c8\u8fdb\u4e00\u6b65\">LayoutLMv3\uff1a\u8ddd\u79bb\u6587\u6863\u7406\u89e3\u8de8\u6a21\u6001\u5bf9\u9f50\u53c8\u8fdb\u4e00\u6b65<\/h2>\n\n\n\n<p>\u73b0\u5b9e\u751f\u6d3b\u4e2d\u7684\u6587\u6863\u4e0d\u4ec5\u6709\u5927\u91cf\u7684\u6587\u672c\u4fe1\u606f\uff0c\u540c\u65f6\u4e5f\u5305\u542b\u4e30\u5bcc\u7684\u5e03\u5c40\u548c\u89c6\u89c9\u4fe1\u606f\uff0c\u5e76\u4e14\u8fd9\u4e09\u79cd\u6a21\u6001\u5728\u6587\u6863\u4e2d\u6709\u5929\u7136\u7684\u5bf9\u9f50\u7279\u6027\u3002\u5982\u4f55\u5bf9\u8fd9\u4e9b\u6587\u6863\u8fdb\u884c\u5efa\u6a21\u5e76\u4e14\u901a\u8fc7\u8bad\u7ec3\u8fbe\u5230\u8de8\u6a21\u6001\u5bf9\u9f50\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u7814\u7a76\u8bfe\u9898\u3002\u5bf9\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728\u6587\u6863\u667a\u80fd\u9886\u57df\u8fdb\u884c\u4e86\u8bf8\u591a\u63a2\u7d22\uff0c\u63a8\u51fa\u4e86\u901a\u7528\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3 LayoutLM \u7cfb\u5217\u7814\u7a76\u6210\u679c\u3002\u8fd9\u4e9b\u6210\u679c\u4e0d\u4ec5\u5728\u5b66\u672f\u754c\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\u548c\u8ba4\u53ef\uff0c\u5728\u5de5\u4e1a\u754c\u4e5f\u5f97\u5230\u4e86\u5e7f\u6cdb\u5e94\u7528\uff0c\u5982\u5fae\u8f6f Azure \u8ba4\u77e5\u670d\u52a1\u4e2d\u7684\u8868\u5355\u8bc6\u522b\u5668\uff08Form Recognizer\uff09\u7b49\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>LayoutLM\u2014\u2014\u901a\u8fc7\u5c06\u6587\u672c\u548c\u5e03\u5c40\u8fdb\u884c\u8054\u5408\u9884\u8bad\u7ec3\uff0c\u5728\u591a\u79cd\u6587\u6863\u7406\u89e3\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u663e\u8457\u63d0\u5347\u3002<\/li>\n\n\n\n<li>LayoutLMv2\u2014\u2014\u901a\u8fc7\u5c06\u89c6\u89c9\u7279\u5f81\u4fe1\u606f\u878d\u5165\u5230\u9884\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u5927\u5927\u63d0\u9ad8\u4e86\u6a21\u578b\u7684\u56fe\u50cf\u7406\u89e3\u80fd\u529b\u3002<\/li>\n\n\n\n<li>LayoutXLM\u2014\u2014\u57fa\u4e8e LayoutLMv2 \u7684\u6a21\u578b\u7ed3\u6784\uff0c\u901a\u8fc7\u4f7f\u752853\u79cd\u8bed\u8a00\u8fdb\u884c\u9884\u8bad\u7ec3\uff0c\u62d3\u5c55\u4e86 LayoutLM \u7684\u591a\u8bed\u8a00\u652f\u6301\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u5982\u4eca\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u53c8\u63d0\u51fa\u4e86\u65b0\u4e00\u4ee3\u7684\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3\u6a21\u578b LayoutLMv3\uff0c\u901a\u8fc7\u7b80\u5355\u7684\u6a21\u578b\u67b6\u6784\u548c\u7edf\u4e00\u7684\u9884\u8bad\u7ec3\u76ee\u6807\uff0c\u5927\u5927\u7b80\u5316\u4e86\u6a21\u578b\u8bbe\u8ba1\uff0c\u589e\u5f3a\u4e86\u4e0b\u6e38\u4efb\u52a1\u7684\u6613\u7528\u6027\u3002\u7814\u7a76\u5458\u4eec\u5206\u522b\u57285000\u4e07\u5f20\u4e2d\u6587\u548c1100\u4e07\u5f20\u82f1\u6587\u6587\u6863\u56fe\u50cf\u4e0a\u9884\u8bad\u7ec3\u4e86 LayoutLMv3 \u6a21\u578b\uff0c\u4ee5\u6ee1\u8db3\u5e7f\u5927\u4e2d\u82f1\u6587\u7528\u6237\u7684\u9700\u6c42\u3002LayoutLMv3 \u5728\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e2d\uff08\u5982\u8868\u5355\u7406\u89e3\u3001\u5c0f\u7968\u7406\u89e3\u548c\u6587\u6863\u89c6\u89c9\u95ee\u7b54\uff09\u548c\u5728\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e2d\uff08\u5982\u6587\u6863\u56fe\u50cf\u5206\u7c7b\u548c\u6587\u6863\u5e03\u5c40\u5206\u6790\uff09\u5747\u6709\u51fa\u8272\u8868\u73b0\u3002<\/p>\n\n\n\n<p>\u76f8\u5173\u8bba\u6587\u88ab\u8ba1\u7b97\u673a\u591a\u5a92\u4f53\u9886\u57df\u9876\u4f1a ACM Multimedia 2022 \u63a5\u6536\u4e3a Oral Presentation\uff08\u53e3\u5934\u62a5\u544a\u8bba\u6587\uff09\uff0c\u4ee3\u7801\u53ca\u6a21\u578b\u5747\u5df2\u5f00\u6e90\u3002\uff08\u8bba\u6587\u53ca\u4ee3\u7801\u94fe\u63a5\uff0c\u8bf7\u89c1\u6587\u672b\uff09<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"layoutlmv3-\u4ee5\u7edf\u4e00\u7684\u6587\u672c\u548c\u56fe\u50cf\u63a9\u7801\u5efa\u6a21\u76ee\u6807\u6765\u9884\u8bad\u7ec3\u591a\u6a21\u6001\u6a21\u578b\">LayoutLMv3 \u4ee5\u7edf\u4e00\u7684\u6587\u672c\u548c\u56fe\u50cf\u63a9\u7801\u5efa\u6a21\u76ee\u6807\u6765\u9884\u8bad\u7ec3\u591a\u6a21\u6001\u6a21\u578b<\/h2>\n\n\n\n<p>\u81ea\u76d1\u7763\u5b66\u4e60\u5229\u7528\u5927\u91cf\u65e0\u6807\u6ce8\u6570\u636e\uff0c\u5728\u8868\u5f81\u5b66\u4e60\u4e2d\u53d6\u5f97\u4e86\u5feb\u901f\u8fdb\u5c55\u3002\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7814\u7a76\u4e2d\uff0cBERT \u9996\u5148\u63d0\u51fa\u4e86\u201c\u63a9\u7801\u8bed\u8a00\u5efa\u6a21\u201d\uff08Masked Language Modeling, MLM\uff09\u7684\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u65b9\u6cd5\uff0c\u901a\u8fc7\u968f\u673a\u906e\u76d6\u6587\u672c\u4e2d\u4e00\u5b9a\u6bd4\u4f8b\u7684\u5355\u8bcd\uff0c\u5e76\u6839\u636e\u4e0a\u4e0b\u6587\u91cd\u5efa\u88ab\u906e\u76d6\u7684\u5355\u8bcd\u6765\u5b66\u4e60\u5177\u6709\u4e0a\u4e0b\u6587\u8bed\u4e49\u7684\u8868\u5f81\u3002\u867d\u7136\u5927\u591a\u6570\u7528\u4e8e\u6587\u6863\u667a\u80fd\u7684\u591a\u6a21\u6001\u9884\u8bad\u7ec3\u6a21\u578b\u5728\u8bed\u8a00\u6a21\u6001\u4e0a\u4f7f\u7528\u4e86 MLM\uff0c\u4f46\u5b83\u4eec\u5728\u56fe\u50cf\u6a21\u6001\u7684\u9884\u8bad\u7ec3\u76ee\u6807\u6709\u6240\u4e0d\u540c\u3002\u4f8b\u5982\uff0c\u6709\u7684\u6a21\u578b\u76ee\u6807\u662f\u91cd\u5efa\u5bc6\u96c6\u7684\u56fe\u50cf\u50cf\u7d20\u6216\u662f\u91cd\u5efa\u8fde\u7eed\u7684\u5c40\u90e8\u533a\u57df\u7279\u5f81\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u503e\u5411\u4e8e\u5b66\u4e60\u5608\u6742\u7684\u7ec6\u8282\uff0c\u800c\u4e0d\u662f\u9ad8\u5c42\u7ed3\u6784\uff08\u5982\u6587\u6863\u5e03\u5c40\uff09\u3002\u5e76\u4e14\uff0c\u56fe\u50cf\u548c\u6587\u672c\u76ee\u6807\u7684\u7c92\u5ea6\u4e0d\u540c\u8fdb\u4e00\u6b65\u589e\u52a0\u4e86\u8de8\u6a21\u6001\u5bf9\u9f50\u5b66\u4e60\u7684\u96be\u5ea6\uff0c\u800c\u8de8\u6a21\u6001\u5bf9\u9f50\u5b66\u4e60\u5bf9\u591a\u6a21\u6001\u8868\u793a\u5b66\u4e60\u975e\u5e38\u5173\u952e\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u514b\u670d\u6587\u672c\u548c\u56fe\u50cf\u5728\u9884\u8bad\u7ec3\u76ee\u6807\u4e0a\u7684\u5dee\u5f02\uff0c\u4fc3\u8fdb\u591a\u6a21\u6001\u8868\u5f81\u5b66\u4e60\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 LayoutLMv3\uff0c\u4ee5\u7edf\u4e00\u7684\u6587\u672c\u548c\u56fe\u50cf\u63a9\u7801\u5efa\u6a21\u76ee\u6807\u6765\u9884\u8bad\u7ec3\u591a\u6a21\u6001\u6a21\u578b\uff0c\u5373 LayoutLMv3 \u5b66\u4e60\u91cd\u5efa\u8bed\u8a00\u6a21\u6001\u7684\u906e\u76d6\u8bcd ID\uff0c\u5e76\u5bf9\u79f0\u5730\u91cd\u5efa\u56fe\u50cf\u6a21\u6001\u7684\u906e\u76d6\u56fe\u50cf\u5757 ID\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"644\" height=\"724\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-2.png\" alt=\"\u56fe2\uff1a\u5173\u4e8e\u56fe\u50cf\u8868\u793a\u65b9\u5f0f\u548c\u9884\u8bad\u7ec3\u76ee\u6807\u7684\u6bd4\u8f83\" class=\"wp-image-1070823\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-2.png 644w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-2-267x300.png 267w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-2-160x180.png 160w\" sizes=\"auto, (max-width: 644px) 100vw, 644px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe2\uff1a\u5173\u4e8e\u56fe\u50cf\u8868\u793a\u65b9\u5f0f\u548c\u9884\u8bad\u7ec3\u76ee\u6807\u7684\u6bd4\u8f83<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u5bf9\u4e8e\u6587\u6863\u6765\u8bf4\uff0c\u6bcf\u4e2a\u6587\u672c\u8bcd\u90fd\u5bf9\u5e94\u7740\u4e00\u4e2a\u56fe\u50cf\u5757\u3002\u4e3a\u4e86\u5b66\u4e60\u8fd9\u79cd\u8de8\u6a21\u6001\u7684\u5bf9\u9f50\u5173\u7cfb\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u63d0\u51fa\u4e86\u4e00\u4e2a\u8bcd\u5757\u5bf9\u9f50\u9884\u8bad\u7ec3\u76ee\u6807\uff0c\u901a\u8fc7\u9884\u6d4b\u4e00\u4e2a\u6587\u672c\u8bcd\u7684\u5bf9\u5e94\u56fe\u50cf\u5757\u662f\u5426\u88ab\u906e\u76d6\uff0c\u6765\u5b66\u4e60\u8fd9\u79cd\u8bed\u8a00\u2014\u2014\u56fe\u50cf\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u5173\u7cfb\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"layoutlmv3-\u6a21\u578b\u67b6\u6784-\u76f4\u63a5\u5229\u7528\u6587\u6863\u56fe\u50cf\u7684\u56fe\u50cf\u5757-\u8282\u7701\u53c2\u6570\u5e76\u907f\u514d\u533a\u57df\u6807\u6ce8\">LayoutLMv3 \u6a21\u578b\u67b6\u6784\uff1a\u76f4\u63a5\u5229\u7528\u6587\u6863\u56fe\u50cf\u7684\u56fe\u50cf\u5757\uff0c\u8282\u7701\u53c2\u6570\u5e76\u907f\u514d\u533a\u57df\u6807\u6ce8<\/h2>\n\n\n\n<p>\u5728\u6a21\u578b\u67b6\u6784\u8bbe\u8ba1\u4e0a\uff0cLayoutLMv3 \u4e0d\u4f9d\u8d56\u590d\u6742\u7684 CNN \u6216 Faster R-CNN \u7f51\u7edc\u6765\u8868\u5f81\u56fe\u50cf\uff0c\u800c\u662f\u76f4\u63a5\u5229\u7528\u6587\u6863\u56fe\u50cf\u7684\u56fe\u50cf\u5757\uff0c\u4ece\u800c\u5927\u5927\u8282\u7701\u4e86\u53c2\u6570\u5e76\u907f\u514d\u4e86\u590d\u6742\u7684\u6587\u6863\u9884\u5904\u7406\uff08\u5982\u4eba\u5de5\u6807\u6ce8\u76ee\u6807\u533a\u57df\u6846\u548c\u6587\u6863\u76ee\u6807\u68c0\u6d4b\uff09\u3002\u7b80\u5355\u7684\u7edf\u4e00\u67b6\u6784\u548c\u8bad\u7ec3\u76ee\u6807\u4f7f LayoutLMv3 \u6210\u4e3a\u901a\u7528\u7684\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u53ef\u9002\u7528\u4e8e\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u548c\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863 AI \u4efb\u52a1\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1020\" height=\"687\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-3.png\" alt=\"\u56fe3\uff1aLayoutLMv3 \u7684\u67b6\u6784\u548c\u9884\u8bad\u7ec3\u76ee\u6807\" class=\"wp-image-1070826\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-3.png 1020w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-3-300x202.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-3-768x517.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-3-240x162.png 240w\" sizes=\"auto, (max-width: 1020px) 100vw, 1020px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe3\uff1aLayoutLMv3 \u7684\u67b6\u6784\u548c\u9884\u8bad\u7ec3\u76ee\u6807<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728\u4e94\u4e2a\u6570\u636e\u96c6\u4e2d\u8bc4\u4f30\u4e86\u9884\u8bad\u7ec3\u7684 LayoutLMv3 \u6a21\u578b\uff0c\u5305\u62ec\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u7684\u6570\u636e\u96c6\uff1a\u8868\u5355\u7406\u89e3 FUNSD \u6570\u636e\u96c6\uff0c\u7968\u636e\u7406\u89e3 CORD \u6570\u636e\u96c6\uff0c\u6587\u6863\u89c6\u89c9\u95ee\u7b54 DocVQA \u6570\u636e\u96c6\uff1b\u4ee5\u53ca\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6570\u636e\u96c6\uff1a\u6587\u6863\u56fe\u50cf\u5206\u7c7b RVL-CDIP \u6570\u636e\u96c6\uff0c\u6587\u6863\u5e03\u5c40\u5206\u6790 PubLayNet \u6570\u636e\u96c6\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cLayoutLMv3 \u5728\u8fd9\u4e9b\u6570\u636e\u96c6\u4e0a\u4ee5\u66f4\u5c11\u7684\u53c2\u6570\u8fbe\u5230\u4e86\u66f4\u4f18\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p>LayoutLMv3 \u8fd8\u5e94\u7528\u4e86\u6587\u672c\u2014\u2014\u56fe\u50cf\u591a\u6a21\u6001 Transformer \u67b6\u6784\u6765\u5b66\u4e60\u8de8\u6a21\u6001\u8868\u5f81\u3002\u6587\u672c\u5411\u91cf\u7531\u8bcd\u5411\u91cf\u3001\u8bcd\u7684\u4e00\u7ef4\u4f4d\u7f6e\u5411\u91cf\u548c\u4e8c\u7ef4\u4f4d\u7f6e\u5411\u91cf\u76f8\u52a0\u5f97\u5230\u3002\u6587\u6863\u56fe\u50cf\u7684\u6587\u672c\u548c\u5176\u76f8\u5e94\u7684\u4e8c\u7ef4\u4f4d\u7f6e\u4fe1\u606f\uff08\u5e03\u5c40\u4fe1\u606f\uff09\u5219\u5229\u7528\u5149\u5b66\u5b57\u7b26\u8bc6\u522b\uff08OCR\uff09\u5de5\u5177\u62bd\u53d6\u3002\u56e0\u4e3a\u6587\u672c\u7684\u90bb\u63a5\u8bcd\u901a\u5e38\u8868\u8fbe\u4e86\u76f8\u4f3c\u7684\u8bed\u4e49\uff0cLayoutLMv3 \u5171\u4eab\u4e86\u90bb\u63a5\u8bcd\u7684\u4e8c\u7ef4\u4f4d\u7f6e\u5411\u91cf\uff0c\u800c LayoutLM \u548c LayoutLMv2 \u7684\u6bcf\u4e2a\u8bcd\u5219\u7528\u4e86\u4e0d\u540c\u7684\u4e8c\u7ef4\u4f4d\u7f6e\u5411\u91cf\u3002<\/p>\n\n\n\n<p>\u56fe\u50cf\u5411\u91cf\u7684\u8868\u793a\u901a\u5e38\u4f9d\u8d56\u4e8e CNN \u62bd\u53d6\u7279\u5f81\u56fe\u7f51\u683c\u7279\u5f81\u6216 Faster R-CNN \u63d0\u53d6\u533a\u57df\u7279\u5f81\uff0c\u8fd9\u4e9b\u65b9\u5f0f\u589e\u52a0\u4e86\u8ba1\u7b97\u5f00\u9500\u6216\u4f9d\u8d56\u4e8e\u533a\u57df\u6807\u6ce8\u3002\u56e0\u6b64\uff0c\u7814\u7a76\u5458\u4eec\u5c06\u56fe\u50cf\u5757\u7ecf\u8fc7\u7ebf\u6027\u6620\u5c04\u83b7\u5f97\u56fe\u50cf\u7279\u5f81\uff0c\u8fd9\u79cd\u56fe\u50cf\u8868\u793a\u65b9\u5f0f\u6700\u65e9\u5728 ViT \u4e2d\u88ab\u63d0\u51fa\uff0c\u8ba1\u7b97\u5f00\u9500\u6781\u5c0f\u4e14\u4e0d\u4f9d\u8d56\u4e8e\u533a\u57df\u6807\u6ce8\uff0c\u6709\u6548\u89e3\u51b3\u4e86\u4ee5\u4e0a\u95ee\u9898\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u9996\u5148\u5c06\u56fe\u50cf\u7f29\u653e\u4e3a\u7edf\u4e00\u7684\u5927\u5c0f\uff08\u4f8b\u5982224&#215;224\uff09\uff0c\u7136\u540e\u5c06\u56fe\u50cf\u5207\u5206\u6210\u56fa\u5b9a\u5927\u5c0f\u7684\u5757\uff08\u4f8b\u598216&#215;16\uff09\uff0c\u5e76\u901a\u8fc7\u7ebf\u6027\u6620\u5c04\u83b7\u5f97\u56fe\u50cf\u7279\u5f81\u5e8f\u5217\uff0c\u518d\u52a0\u4e0a\u53ef\u5b66\u4e60\u7684\u4e00\u7ef4\u4f4d\u7f6e\u5411\u91cf\u540e\u5f97\u5230\u56fe\u50cf\u5411\u91cf\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"layoutlmv3-\u901a\u8fc7\u4e09\u4e2a\u9884\u8bad\u7ec3\u76ee\u6807\u4ee5\u81ea\u76d1\u7763\u7684\u65b9\u5f0f\u5b66\u4e60\u591a\u6a21\u6001\u8868\u793a\">LayoutLMv3 \u901a\u8fc7\u4e09\u4e2a\u9884\u8bad\u7ec3\u76ee\u6807\u4ee5\u81ea\u76d1\u7763\u7684\u65b9\u5f0f\u5b66\u4e60\u591a\u6a21\u6001\u8868\u793a<\/h2>\n\n\n\n<p>\u63a9\u7801\u8bed\u8a00\u5efa\u6a21\uff08Masked Language Modeling, MLM\uff09\u3002\u4e3a\u4e86\u5229\u4e8e\u6a21\u578b\u5b66\u4e60\u5e03\u5c40\u4fe1\u606f\u4e0e\u6587\u672c\u548c\u56fe\u50cf\u4e4b\u95f4\u7684\u5bf9\u5e94\u5173\u7cfb\uff0c\u8be5\u4efb\u52a1\u968f\u673a\u906e\u76d630%\u7684\u6587\u672c\u8bcd\u5411\u91cf\uff0c\u4f46\u4fdd\u7559\u5bf9\u5e94\u7684\u4e8c\u7ef4\u4f4d\u7f6e\uff08\u5e03\u5c40\uff09\u4fe1\u606f\u3002\u7c7b\u4f3c BERT \u548c LayoutLM\uff0c\u6a21\u578b\u76ee\u6807\u662f\u6839\u636e\u672a\u88ab\u906e\u76d6\u7684\u56fe\u6587\u548c\u5e03\u5c40\u4fe1\u606f\u8fd8\u539f\u6587\u672c\u4e2d\u88ab\u906e\u76d6\u7684\u8bcd\u3002<\/p>\n\n\n\n<p>\u63a9\u7801\u56fe\u50cf\u5efa\u6a21\uff08Masked Image Modeling, MIM\uff09\u3002\u4e3a\u4e86\u9f13\u52b1\u6a21\u578b\u4ece\u6587\u672c\u548c\u56fe\u50cf\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u63a8\u6d4b\u56fe\u50cf\u4fe1\u606f\uff0c\u8be5\u4efb\u52a1\u968f\u673a\u906e\u76d6\u4e86\u7ea640%\u7684\u56fe\u50cf\u5757\u3002\u7c7b\u4f3c BEiT\uff0c\u6a21\u578b\u76ee\u6807\u662f\u6839\u636e\u672a\u88ab\u906e\u76d6\u7684\u6587\u672c\u548c\u56fe\u50cf\u7684\u4fe1\u606f\u8fd8\u539f\u88ab\u906e\u76d6\u7684\u56fe\u50cf\u5757\u7ecf\u8fc7\u79bb\u6563\u5316\u7684 ID\u3002<\/p>\n\n\n\n<p>\u8bcd\u5757\u5bf9\u9f50\uff08Word-Patch Alignment, WPA\uff09\u3002\u5bf9\u4e8e\u6587\u6863\u6765\u8bf4\uff0c\u6bcf\u4e2a\u6587\u672c\u8bcd\u90fd\u5bf9\u5e94\u7740\u4e00\u4e2a\u56fe\u50cf\u5757\u3002\u7531\u4e8e\u524d\u4e24\u4e2a\u4efb\u52a1\u968f\u673a\u906e\u76d6\u4e86\u90e8\u5206\u6587\u672c\u8bcd\u548c\u56fe\u50cf\u5757\uff0c\u6a21\u578b\u65e0\u6cd5\u663e\u5f0f\u5730\u5b66\u4e60\u8fd9\u79cd\u6587\u672c\u8bcd\u548c\u56fe\u50cf\u5757\u4e4b\u95f4\u7684\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u5173\u7cfb\u3002\u8be5\u76ee\u6807\u901a\u8fc7\u663e\u5f0f\u5730\u9884\u6d4b\u4e00\u4e2a\u6587\u672c\u8bcd\u7684\u5bf9\u5e94\u56fe\u50cf\u5757\u662f\u5426\u88ab\u63a9\u76d6\u6765\u5b66\u4e60\u8bed\u8a00\u548c\u89c6\u89c9\u6a21\u6001\u4e4b\u95f4\u7684\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u5173\u7cfb\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u5b9e\u9a8c\u53ca\u7ed3\u679c\">\u5b9e\u9a8c\u53ca\u7ed3\u679c<\/h2>\n\n\n\n<p>\u4e3a\u4e86\u5b66\u4e60\u5404\u79cd\u6587\u6863\u4efb\u52a1\u7684\u901a\u7528\u8868\u793a\uff0cLayoutLMv3 \u7684\u9884\u8bad\u7ec3\u6570\u636e\u96c6\u662f IIT-CDIP\uff0c\u5305\u542b\u5927\u7ea61100\u4e07\u5f20\u626b\u63cf\u6587\u6863\u56fe\u50cf\u3002\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8bad\u7ec3\u4e86 BASE\u3001LARGE \u4e24\u79cd\u89c4\u6a21\u7684\u6a21\u578b\uff0c\u53c2\u6570\u91cf\u5206\u522b\u662f133M\u3001368M\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"1-\u5fae\u8c03\u591a\u6a21\u6001\u4efb\u52a1-\u8fdc\u8d85-sota\">1. \u5fae\u8c03\u591a\u6a21\u6001\u4efb\u52a1\uff1a\u8fdc\u8d85 SOTA<\/h3>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u5728\u56db\u4e2a\u591a\u6a21\u6001\u4efb\u52a1\u4e0a\u5bf9 LayoutLMv3 \u8fdb\u884c\u4e86\u5fae\u8c03\uff1a<\/p>\n\n\n\n<p>\uff081\uff09\u8868\u5355\u7406\u89e3\u4efb\u52a1\u5bf9\u8868\u5355\u7684\u6587\u672c\u5185\u5bb9\u8fdb\u884c\u5e8f\u5217\u6807\u6ce8\u3002FUNSD \u662f\u4e00\u4e2a\u6587\u6863\u626b\u63cf\u8868\u5355\u7406\u89e3\u6570\u636e\u96c6\uff0c\u5305\u542b199\u4e2a\u6587\u6863\uff0c\u67099,707\u4e2a\u8bed\u4e49\u5b9e\u4f53\u7684\u6807\u6ce8\u3002FUNSD \u6570\u636e\u96c6\u7684\u8bed\u4e49\u5b9e\u4f53\u6807\u7b7e\u4efb\u52a1\u65e8\u5728\u4e3a\u6bcf\u4e2a\u8bed\u4e49\u5b9e\u4f53\u5206\u7c7b\u4e3a\u201c\u95ee\u9898\u201d\u3001\u201c\u7b54\u6848\u201d\u3001\u201c\u6807\u9898\u201d\u6216\u201c\u5176\u4ed6\u201d\u4e2d\u7684\u4e00\u4e2a\u3002<\/p>\n\n\n\n<p>\uff082\uff09\u7968\u636e\u7406\u89e3\u4efb\u52a1\u9700\u8981\u63d0\u53d6\u7968\u636e\u4fe1\u606f\uff0c\u5bf9\u6bcf\u4e2a\u8bcd\u8fdb\u884c\u8bed\u4e49\u6807\u7b7e\u5206\u7c7b\u3002\u8be5\u4efb\u52a1\u7684\u6570\u636e\u96c6\u662f CORD\uff0c\u5b83\u5305\u542b1,000\u5f20\u6536\u636e\uff0c\u57284\u4e2a\u7c7b\u522b\u4e0b\u5b9a\u4e49\u4e8630\u4e2a\u8bed\u4e49\u6807\u7b7e\u3002<\/p>\n\n\n\n<p>\uff083\uff09\u6587\u6863\u56fe\u50cf\u5206\u7c7b\u4efb\u52a1\u7684\u76ee\u7684\u662f\u9884\u6d4b\u6587\u6863\u56fe\u50cf\u7684\u7c7b\u522b\u3002\u8be5\u4efb\u52a1\u5728 RVL-CDIP \u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u3002RVL-CDIP \u6570\u636e\u96c6\u5305\u542b400,000\u5f20\u6587\u6863\u56fe\u50cf\uff0c\u6807\u670916\u4e2a\u7c7b\u522b\u3002<\/p>\n\n\n\n<p>\uff084\uff09\u6587\u6863\u89c6\u89c9\u95ee\u7b54\u4efb\u52a1\u9700\u8981\u6a21\u578b\u5c06\u6587\u6863\u56fe\u50cf\u548c\u95ee\u9898\u4f5c\u4e3a\u8f93\u5165\uff0c\u5e76\u8f93\u51fa\u4e00\u4e2a\u7b54\u6848\u3002\u8be5\u4efb\u52a1\u5728DocVQA\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u3002DocVQA \u7684\u8bad\u7ec3\u96c6\u5305\u542b\u7ea6\u4e00\u4e07\u5f20\u56fe\u7247\u548c\u56db\u4e07\u4e2a\u95ee\u9898\u3002<\/p>\n\n\n\n<p>\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e0a\uff0cLayoutLMv3 \u53d6\u5f97\u4e86\u6bd4\u4ee5\u524d\u7684\u5de5\u4f5c\u66f4\u597d\u6216\u76f8\u5f53\u7684\u7ed3\u679c\u3002\u4f8b\u5982\uff0c\u5bf9\u4e8eLARGE\u6a21\u578b\u89c4\u6a21\uff0cLayoutLMv3 \u5728 FUNSD \u6570\u636e\u96c6\u4e0a\u53d6\u5f97\u4e8692.08\u7684 F1 \u5206\u6570\uff0c\u5927\u5927\u8d85\u8fc7\u4e86\u6b64\u524d LARGE \u89c4\u6a21\u7684 SOTA \u7ed3\u679c\uff0885.14\uff09\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"578\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-1024x578.png\" alt=\"\u88681\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 CORD\u3001FUNSD\u3001RVL-CDIP\u3001DocVQA \u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4\" class=\"wp-image-1070832\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-1024x578.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-300x169.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-768x433.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-240x135.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4-640x360.png 640w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-4.png 1179w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88681\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 CORD\u3001FUNSD\u3001RVL-CDIP\u3001DocVQA \u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"2-\u5fae\u8c03\u89c6\u89c9\u4efb\u52a1-\u6240\u6709\u6307\u6807\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\">2. \u5fae\u8c03\u89c6\u89c9\u4efb\u52a1\uff1a\u6240\u6709\u6307\u6807\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b<\/h3>\n\n\n\n<p>\u4e3a\u4e86\u8bc1\u660e\u591a\u6a21\u6001\u9884\u8bad\u7ec3\u6a21\u578b LayoutLMv3 \u5728\u89c6\u89c9\u9886\u57df\u5e94\u7528\u7684\u901a\u7528\u6027\uff0c\u7814\u7a76\u5458\u4eec\u5c06 LayoutLMv3 \u5e94\u7528\u4e8e\u4e86\u6587\u6863\u5e03\u5c40\u5206\u6790\u4efb\u52a1\u3002\u6587\u6863\u5e03\u5c40\u5206\u6790\u4efb\u52a1\u65e8\u5728\u4e3a\u6587\u6863\u7684\u4e0d\u540c\u5e03\u5c40\u5143\u7d20\u68c0\u6d4b\u4e8c\u7ef4\u4f4d\u7f6e\u548c\u7c7b\u522b\u3002\u8fd9\u9879\u4efb\u52a1\u6709\u52a9\u4e8e\u5c06\u6587\u6863\u89e3\u6790\u6210\u673a\u5668\u53ef\u8bfb\u7684\u683c\u5f0f\uff0c\u4f9b\u4e0b\u6e38\u5e94\u7528\u7a0b\u5e8f\u4f7f\u7528\u3002\u7814\u7a76\u5458\u4eec\u5c06\u8fd9\u9879\u4efb\u52a1\u5efa\u6a21\u4e3a\u76ee\u6807\u68c0\u6d4b\u95ee\u9898\uff0c\u4ee5 LayoutLMv3 \u4e3a\u7279\u5f81\u63d0\u53d6\u7f51\u7edc\uff0c\u5c06\u4e0d\u540c\u5c42\u63d0\u53d6\u7684\u7279\u5f81\u6574\u5408\u5230\u76ee\u6807\u68c0\u6d4b\u5668\u4e2d\u3002\u7814\u7a76\u5458\u4eec\u5728 PubLayNet \u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u4e86\u5b9e\u9a8c\u3002\u8be5\u6570\u636e\u96c6\u5305\u542b\u4e09\u5341\u591a\u4e07\u5f20\u7814\u7a76\u8bba\u6587\u56fe\u50cf\uff0c\u6bcf\u5f20\u56fe\u50cf\u90fd\u6709\u5e03\u5c40\u8fb9\u754c\u6846\u7684\u4f4d\u7f6e\u548c\u7c7b\u522b\u6807\u6ce8\uff0c\u7c7b\u522b\u6db5\u76d6\u6587\u672c\u3001\u6807\u9898\u3001\u5217\u8868\u3001\u56fe\u548c\u8868\u3002\u4e0e\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u548c\u540c\u671f\u7684\u89c6\u89c9 Transformer \u9884\u8bad\u7ec3\u6a21\u578b\u76f8\u6bd4\uff0cLayoutLMv3 \u5728\u6240\u6709\u6307\u6807\u4e0a\u90fd\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"203\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5-1024x203.png\" alt=\"\u88682\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 PubLayNet \u6570\u636e\u96c6\u5173\u4e8e\u6587\u6863\u5e03\u5c40\u5206\u6790\u4efb\u52a1\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4\" class=\"wp-image-1070835\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5-1024x203.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5-300x60.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5-768x152.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5-240x48.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-5.png 1114w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88682\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 PubLayNet \u6570\u636e\u96c6\u5173\u4e8e\u6587\u6863\u5e03\u5c40\u5206\u6790\u4efb\u52a1\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"3-\u6d88\u878d\u5b9e\u9a8c-\u6709\u529b\u8bc1\u660e\u4e86-layoutlmv3-\u9002\u5408\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863\u5206\u6790\u4efb\u52a1\">3. \u6d88\u878d\u5b9e\u9a8c\uff1a\u6709\u529b\u8bc1\u660e\u4e86 LayoutLMv3 \u9002\u5408\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863\u5206\u6790\u4efb\u52a1<\/h3>\n\n\n\n<p>\u4e3a\u4e86\u7814\u7a76 LayoutLMv3 \u7684\u56fe\u50cf\u8868\u793a\u65b9\u6cd5\u548c\u9884\u8bad\u7ec3\u76ee\u6807\u7684\u6548\u679c\uff0c\u7814\u7a76\u5458\u4eec\u5728\u56db\u4e2a\u5178\u578b\u7684\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u4e86\u6d88\u878d\u5b9e\u9a8c\u3002\u57fa\u7840\u6a21\u578b\u53ea\u4f7f\u7528\u6587\u672c\u548c\u5e03\u5c40\u4fe1\u606f\uff0c\u7528 MLM \u76ee\u6807\u8fdb\u884c\u9884\u8bad\u7ec3\u3002\u63a5\u7740\uff0c\u7814\u7a76\u5458\u4eec\u4f9d\u6b21\u4e3a\u57fa\u7840\u6a21\u578b\u589e\u52a0\u56fe\u50cf\u8868\u793a\uff0cMIM \u548c WPA \u9884\u8bad\u7ec3\u76ee\u6807\u3002<\/p>\n\n\n\n<p>\u901a\u8fc7\u8868\u683c\u4e2d\u7684\u7ed3\u679c\uff0c\u53ef\u89c2\u5bdf\u5230\uff0c\u57fa\u7840\u6a21\u578b\u6ca1\u6709\u56fe\u50cf\u8868\u793a\uff0c\u4e0d\u80fd\u8fdb\u884c\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863\u5206\u6790\u4efb\u52a1\u3002LayoutLMv3 \u91c7\u7528\u56fe\u50cf\u5757\u7684\u7ebf\u6027\u6620\u5c04\u4f5c\u4e3a\u56fe\u50cf\u8868\u793a\uff0c\u8bbe\u8ba1\u7b80\u5355\uff0c\u53ea\u5f15\u5165\u4e860.6M\u7684\u53c2\u6570\u3002\u8fd9\u79cd\u56fe\u50cf\u8868\u793a\u901a\u8fc7\u4e0e MIM \u9884\u8bad\u7ec3\u76ee\u6807\u7ed3\u5408\uff0c\u4e0d\u4ec5\u80fd\u652f\u6301\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\uff0c\u8fd8\u80fd\u63d0\u5347\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u7684\u8868\u73b0\u3002\u7ed3\u5408 WPA \u9884\u8bad\u7ec3\u76ee\u6807\u53c8\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u6240\u6709\u4e0b\u6e38\u4efb\u52a1\u7684\u6027\u80fd\uff0c\u8bc1\u5b9e\u4e86 WPA \u5728\u8de8\u6a21\u6001\u8868\u793a\u5b66\u4e60\u548c\u56fe\u50cf\u8868\u793a\u5b66\u4e60\u4e2d\u7684\u6709\u6548\u6027\uff0c\u5f3a\u8c03\u4e86\u89c6\u89c9\u548c\u8bed\u8a00\u8de8\u6a21\u6001\u5bf9\u9f50\u5b66\u4e60\u7684\u91cd\u8981\u6027\u3002\u6b64\u5916\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u53d1\u73b0 MIM \u53ef\u4ee5\u5e2e\u52a9\u89c4\u8303\u5316\u8bad\u7ec3\uff0c\u5bf9\u4e8e PubLayNet \u4e0a\u7684\u6587\u6863\u5e03\u5c40\u5206\u6790\u7b49\u89c6\u89c9\u4efb\u52a1\u7684\u6a21\u578b\u6536\u655b\u81f3\u5173\u91cd\u8981\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"203\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6-1024x203.png\" alt=\"\u88683\uff1a\u5173\u4e8e\u56fe\u50cf\u8868\u5f81\u548c\u9884\u8bad\u7ec3\u76ee\u6807\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4\" class=\"wp-image-1070838\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6-1024x203.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6-300x59.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6-768x152.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6-240x47.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-6.png 1026w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88683\uff1a\u5173\u4e8e\u56fe\u50cf\u8868\u5f81\u548c\u9884\u8bad\u7ec3\u76ee\u6807\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"layoutlmv3-\u7528\u4e8e\u4e2d\u6587\u6587\u6863\u7406\u89e3\u4e5f\u6709\u51fa\u8272\u8868\u73b0\">LayoutLMv3 \u7528\u4e8e\u4e2d\u6587\u6587\u6863\u7406\u89e3\u4e5f\u6709\u51fa\u8272\u8868\u73b0<\/h2>\n\n\n\n<p>\u4ee5\u4e0a\u63cf\u8ff0\u7684 LayoutLMv3 \u6a21\u578b\u7684\u9884\u8bad\u7ec3\u6570\u636e\u5747\u4e3a\u82f1\u8bed\uff0c\u53ef\u5728\u82f1\u8bed\u7684\u4e0b\u6e38\u4efb\u52a1\u6570\u636e\u96c6\u4e0a\u5fae\u8c03\u3002\u4e3a\u4e86\u6ee1\u8db3\u5e7f\u5927\u4e2d\u6587\u7528\u6237\u7684\u9700\u6c42\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u8bad\u7ec3\u4e86\u4e2d\u6587\u7684 LayoutLMv3 \u6a21\u578b\uff0c\u53ef\u7528\u4e8e\u4e2d\u6587\u4efb\u52a1\u3002<\/p>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u901a\u8fc7\u5927\u89c4\u6a21\u6570\u636e\u6293\u53d6\u83b7\u5f97\u4e86\u6d77\u91cf\u7684\u6570\u5b57\u6587\u6863\uff0c\u5e76\u7528\u81ea\u52a8\u5316\u8bed\u8a00\u68c0\u6d4b\u5de5\u5177\u7b5b\u9009\u51fa\u4e865000\u4e07\u5f20\u4e2d\u6587\u6587\u6863\u56fe\u50cf\u7528\u4e8e\u9884\u8bad\u7ec3\u3002\u7814\u7a76\u5458\u4eec\u5728\u4e00\u4e2a\u4e2d\u6587\u89c6\u89c9\u4fe1\u606f\u62bd\u53d6\u4efb\u52a1\u4e0a\u9a8c\u8bc1\u4e86\u6a21\u578b\u7684\u6548\u679c\u3002\u7531\u88684\u53ef\u89c1\uff0cLayoutLMv3 \u5728\u6240\u6709\u7c7b\u522b\u5e73\u5747\u5206\u6570\u4e0a\u53d6\u5f97\u4e8699.21\u7684\u9ad8\u5206\uff0c\u663e\u8457\u8d85\u8fc7\u4e86\u5176\u4ed6\u6a21\u578b\uff0c\u9a8c\u8bc1\u4e86 LayoutLMv3 \u4e2d\u6587\u6a21\u578b\u7684\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"190\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-1024x190.png\" alt=\"\u88684\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 EPHOIE \u4e2d\u6587\u6570\u636e\u96c6\u5173\u4e8e\u89c6\u89c9\u4fe1\u606f\u62bd\u53d6\u4efb\u52a1\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4\" class=\"wp-image-1070841\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-1024x190.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-300x56.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-768x142.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-1536x285.png 1536w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-2048x380.png 2048w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/layoutlmv3-7-240x44.png 240w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88684\uff1aLayoutLMv3 \u548c\u5df2\u6709\u5de5\u4f5c\u5728 EPHOIE \u4e2d\u6587\u6570\u636e\u96c6\u5173\u4e8e\u89c6\u89c9\u4fe1\u606f\u62bd\u53d6\u4efb\u52a1\u7684\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5927\u91cf\u7684\u5b9e\u9a8c\u7ed3\u679c\u90fd\u8bc1\u660e\u4e86 LayoutLMv3 \u7684\u901a\u7528\u6027\u548c\u4f18\u8d8a\u6027\uff0c\u5b83\u4e0d\u4ec5\u9002\u7528\u4e8e\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u548c\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863\u667a\u80fd\u4efb\u52a1\uff0c\u8fd8\u53ef\u4ee5\u4ee5\u66f4\u5c11\u7684\u53c2\u6570\u83b7\u5f97\u66f4\u597d\u6216\u76f8\u5f53\u7684\u6027\u80fd\u3002\u672a\u6765\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5c06\u63a2\u7d22\u6269\u5927\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u89c4\u6a21\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u5229\u7528\u66f4\u591a\u7684\u8bad\u7ec3\u6570\u636e\u6765\u8fdb\u4e00\u6b65\u63a8\u52a8\u7ed3\u679c\uff1b\u540c\u65f6\u6269\u5c55\u6a21\u578b\u7684\u5c11\u6837\u672c\u5b66\u4e60\u80fd\u529b\uff0c\u4ece\u800c\u4fc3\u8fdb\u6a21\u578b\u5728\u6587\u6863\u667a\u80fd\u884c\u4e1a\u4e2d\u66f4\u591a\u4e1a\u52a1\u573a\u666f\u4e0b\u7684\u5e94\u7528\u3002<\/p>\n\n\n\n<p>LayoutLMv3 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2204.08387.pdf<\/p>\n\n\n\n<p>LayoutLMv3 \u4ee3\u7801\u94fe\u63a5\uff1ahttps:\/\/aka.ms\/layoutlmv3<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7f16\u8005\u6309\uff1a\u4f01\u4e1a\u6570\u5b57\u5316\u8f6c\u578b\u4e2d\uff0c\u4ee5\u6587\u6863\u3001\u56fe\u50cf\u7b49\u591a\u6a21\u6001\u5f62\u5f0f\u4e3a\u8f7d\u4f53\u7684\u7ed3\u6784\u5316\u5206\u6790\u548c\u5185\u5bb9\u63d0\u53d6\u662f\u5176\u4e2d\u7684\u5173\u952e\u4e00\u73af\uff0c\u5feb\u901f\u3001\u81ea\u52a8\u3001\u7cbe\u51c6\u5730\u5904\u7406\u5305\u62ec\u5408\u540c\u3001\u7968\u636e\u3001\u62a5\u544a\u7b49\u4fe1\u606f\uff0c\u5bf9\u63d0\u5347\u73b0\u4ee3\u4f01\u4e1a\u751f\u4ea7\u6548\u7387\u81f3\u5173\u91cd\u8981\u3002\u56e0\u6b64\uff0c\u6587\u6863\u667a\u80fd\u6280\u672f\u5e94\u8fd0\u800c\u751f\u3002\u8fc7\u53bb\u51e0\u5e74\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u63a8\u51fa\u4e86\u901a\u7528\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3 LayoutLM \u7cfb\u5217\u7814\u7a76\u6210\u679c\uff0c\u5e76\u4e0d\u65ad\u4f18\u5316\u6a21\u578b\u5bf9\u6587\u6863\u4e2d\u6587\u672c\u3001\u5e03\u5c40\u548c\u89c6\u89c9\u4fe1\u606f\u7684\u9884\u8bad\u7ec3\u6027\u80fd\u3002\u8fd1\u671f\u53d1\u8868\u7684\u6700\u65b0\u7684 LayoutLM 3.0 \u7248\u672c\uff0c\u5728\u4ee5\u6587\u672c\u548c\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e0a\u6709\u4e86\u66f4\u52a0\u51fa\u8272\u7684\u8868\u73b0\uff0c\u8ba9\u6587\u6863\u7406\u89e3\u6a21\u578b\u5411\u8de8\u6a21\u6001\u5bf9\u9f50\u8fc8\u51fa\u4e00\u5927\u6b65\uff01 \u968f\u7740\u5404\u884c\u5404\u4e1a\u7684\u6570\u5b57\u5316\u8f6c\u578b\uff0c\u6db5\u76d6\u8868\u5355\u3001\u7968\u636e\u3001\u90ae\u4ef6\u3001\u5408\u540c\u3001\u62a5\u544a\u3001\u8bba\u6587\u7b49\u7684\u7535\u5b50\u6587\u6863\u6570\u91cf\u6301\u7eed\u589e\u957f\u3002\u7535\u5b50\u6587\u6863\u5305\u542b\u5927\u91cf\u4e0e\u884c\u4e1a\u76f8\u5173\u7684\u56fe\u50cf\u548c\u6587\u672c\u4fe1\u606f\uff0c\u4eba\u5de5\u5904\u7406\u8fd9\u4e9b\u5927\u91cf\u7684\u4fe1\u606f\u8017\u65f6\u957f\u3001\u6210\u672c\u5927\u3002\u7535\u5b50\u6587\u6863\u7684\u81ea\u52a8\u8bc6\u522b\u3001\u7406\u89e3\u548c\u5206\u6790\u6280\u672f\u5bf9\u63d0\u9ad8\u4e2a\u4eba\u6216\u4f01\u4e1a\u7684\u751f\u4ea7\u529b\u5341\u5206\u91cd\u8981\uff0c\u56e0\u6b64\u6587\u6863\u667a\u80fd\u6280\u672f\u5e94\u8fd0\u800c\u751f\u3002\u6587\u6863\u667a\u80fd\u5229\u7528\u8ba1\u7b97\u673a\u81ea\u52a8\u8bc6\u522b\u3001\u7406\u89e3\u53ca\u5206\u6790\u7535\u5b50\u6587\u6863\uff0c\u5927\u5927\u63d0\u5347\u4e86\u4e2a\u4eba\u548c\u4f01\u4e1a\u5904\u7406\u7535\u5b50\u6587\u6863\u7684\u751f\u4ea7\u529b\uff0c\u662f\u81ea\u7136\u8bed\u8a00\u5904\u7406\u548c\u8ba1\u7b97\u673a\u89c6\u89c9\u4ea4\u53c9\u9886\u57df\u7684\u4e00\u4e2a\u91cd\u8981\u7814\u7a76\u65b9\u5411\u3002 \u867d\u7136\u4e3a\u7279\u5b9a\u4efb\u52a1\u8bbe\u8ba1\u7684\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u9488\u5bf9\u67d0\u4e00\u9879\u6587\u6863\u7406\u89e3\u4efb\u52a1\u80fd\u53d6\u5f97\u8f83\u597d\u7684\u6027\u80fd\uff0c\u4f46\u8fd9\u4e9b\u65b9\u6cd5\u901a\u5e38\u4f9d\u8d56\u4e8e\u6709\u9650\u7684\u6807\u6ce8\u6570\u636e\uff0c\u800c\u5bf9\u4e8e\u6587\u6863\u7406\u89e3\u4efb\u52a1\u5c24\u5176\u662f\u5176\u4e2d\u7684\u4fe1\u606f\u62bd\u53d6\u4efb\u52a1\u6765\u8bf4\uff0c\u83b7\u53d6\u8be6\u7ec6\u7684\u6807\u6ce8\u6570\u636e\u6602\u8d35\u4e14\u8017\u65f6\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u4eba\u5458\u5c06\u76ee\u5149\u8f6c\u5411\u4e86\u90a3\u4e9b\u88ab\u5ffd\u89c6\u7684\u65e0\u6807\u6ce8\u6570\u636e\u4e0a\uff0c\u901a\u8fc7\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u6280\u672f\u6765\u5229\u7528\u73b0\u5b9e\u751f\u6d3b\u4e2d\u5927\u91cf\u7684\u65e0\u6807\u6ce8\u6570\u636e\u3002\u7531\u4e8e\u8fd1\u5e74\u6765\u9884\u8bad\u7ec3\u5728\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u88ab\u5e7f\u6cdb\u5e94\u7528\uff0c\u6240\u4ee5\u8be5\u6280\u672f\u5728\u6587\u6863\u667a\u80fd\u9886\u57df\u4e5f\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\u3002\u7ecf\u8fc7\u9884\u8bad\u7ec3\u7684\u6587\u6863\u667a\u80fd\u6a21\u578b\u53ef\u4ee5\u89e3\u6790\u5e76\u63d0\u53d6\u6587\u6863\u7684\u5404\u7c7b\u4fe1\u606f\uff0c\u8fd9\u5bf9\u6587\u6863\u667a\u80fd\u7684\u5b66\u672f\u7814\u7a76\u548c\u751f\u4ea7\u5e94\u7528\u90fd\u6709\u91cd\u8981\u610f\u4e49\u3002 \u73b0\u5b9e\u751f\u6d3b\u4e2d\u7684\u6587\u6863\u4e0d\u4ec5\u6709\u5927\u91cf\u7684\u6587\u672c\u4fe1\u606f\uff0c\u540c\u65f6\u4e5f\u5305\u542b\u4e30\u5bcc\u7684\u5e03\u5c40\u548c\u89c6\u89c9\u4fe1\u606f\uff0c\u5e76\u4e14\u8fd9\u4e09\u79cd\u6a21\u6001\u5728\u6587\u6863\u4e2d\u6709\u5929\u7136\u7684\u5bf9\u9f50\u7279\u6027\u3002\u5982\u4f55\u5bf9\u8fd9\u4e9b\u6587\u6863\u8fdb\u884c\u5efa\u6a21\u5e76\u4e14\u901a\u8fc7\u8bad\u7ec3\u8fbe\u5230\u8de8\u6a21\u6001\u5bf9\u9f50\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u7814\u7a76\u8bfe\u9898\u3002\u5bf9\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728\u6587\u6863\u667a\u80fd\u9886\u57df\u8fdb\u884c\u4e86\u8bf8\u591a\u63a2\u7d22\uff0c\u63a8\u51fa\u4e86\u901a\u7528\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3 LayoutLM \u7cfb\u5217\u7814\u7a76\u6210\u679c\u3002\u8fd9\u4e9b\u6210\u679c\u4e0d\u4ec5\u5728\u5b66\u672f\u754c\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\u548c\u8ba4\u53ef\uff0c\u5728\u5de5\u4e1a\u754c\u4e5f\u5f97\u5230\u4e86\u5e7f\u6cdb\u5e94\u7528\uff0c\u5982\u5fae\u8f6f Azure \u8ba4\u77e5\u670d\u52a1\u4e2d\u7684\u8868\u5355\u8bc6\u522b\u5668\uff08Form Recognizer\uff09\u7b49\u3002 \u5982\u4eca\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u53c8\u63d0\u51fa\u4e86\u65b0\u4e00\u4ee3\u7684\u6587\u6863\u7406\u89e3\u9884\u8bad\u7ec3\u6a21\u578b LayoutLMv3\uff0c\u901a\u8fc7\u7b80\u5355\u7684\u6a21\u578b\u67b6\u6784\u548c\u7edf\u4e00\u7684\u9884\u8bad\u7ec3\u76ee\u6807\uff0c\u5927\u5927\u7b80\u5316\u4e86\u6a21\u578b\u8bbe\u8ba1\uff0c\u589e\u5f3a\u4e86\u4e0b\u6e38\u4efb\u52a1\u7684\u6613\u7528\u6027\u3002\u7814\u7a76\u5458\u4eec\u5206\u522b\u57285000\u4e07\u5f20\u4e2d\u6587\u548c1100\u4e07\u5f20\u82f1\u6587\u6587\u6863\u56fe\u50cf\u4e0a\u9884\u8bad\u7ec3\u4e86 LayoutLMv3 \u6a21\u578b\uff0c\u4ee5\u6ee1\u8db3\u5e7f\u5927\u4e2d\u82f1\u6587\u7528\u6237\u7684\u9700\u6c42\u3002LayoutLMv3 \u5728\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e2d\uff08\u5982\u8868\u5355\u7406\u89e3\u3001\u5c0f\u7968\u7406\u89e3\u548c\u6587\u6863\u89c6\u89c9\u95ee\u7b54\uff09\u548c\u5728\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u4efb\u52a1\u4e2d\uff08\u5982\u6587\u6863\u56fe\u50cf\u5206\u7c7b\u548c\u6587\u6863\u5e03\u5c40\u5206\u6790\uff09\u5747\u6709\u51fa\u8272\u8868\u73b0\u3002 \u76f8\u5173\u8bba\u6587\u88ab\u8ba1\u7b97\u673a\u591a\u5a92\u4f53\u9886\u57df\u9876\u4f1a ACM Multimedia 2022 \u63a5\u6536\u4e3a Oral Presentation\uff08\u53e3\u5934\u62a5\u544a\u8bba\u6587\uff09\uff0c\u4ee3\u7801\u53ca\u6a21\u578b\u5747\u5df2\u5f00\u6e90\u3002\uff08\u8bba\u6587\u53ca\u4ee3\u7801\u94fe\u63a5\uff0c\u8bf7\u89c1\u6587\u672b\uff09 \u81ea\u76d1\u7763\u5b66\u4e60\u5229\u7528\u5927\u91cf\u65e0\u6807\u6ce8\u6570\u636e\uff0c\u5728\u8868\u5f81\u5b66\u4e60\u4e2d\u53d6\u5f97\u4e86\u5feb\u901f\u8fdb\u5c55\u3002\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7814\u7a76\u4e2d\uff0cBERT \u9996\u5148\u63d0\u51fa\u4e86\u201c\u63a9\u7801\u8bed\u8a00\u5efa\u6a21\u201d\uff08Masked Language Modeling, MLM\uff09\u7684\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u65b9\u6cd5\uff0c\u901a\u8fc7\u968f\u673a\u906e\u76d6\u6587\u672c\u4e2d\u4e00\u5b9a\u6bd4\u4f8b\u7684\u5355\u8bcd\uff0c\u5e76\u6839\u636e\u4e0a\u4e0b\u6587\u91cd\u5efa\u88ab\u906e\u76d6\u7684\u5355\u8bcd\u6765\u5b66\u4e60\u5177\u6709\u4e0a\u4e0b\u6587\u8bed\u4e49\u7684\u8868\u5f81\u3002\u867d\u7136\u5927\u591a\u6570\u7528\u4e8e\u6587\u6863\u667a\u80fd\u7684\u591a\u6a21\u6001\u9884\u8bad\u7ec3\u6a21\u578b\u5728\u8bed\u8a00\u6a21\u6001\u4e0a\u4f7f\u7528\u4e86 MLM\uff0c\u4f46\u5b83\u4eec\u5728\u56fe\u50cf\u6a21\u6001\u7684\u9884\u8bad\u7ec3\u76ee\u6807\u6709\u6240\u4e0d\u540c\u3002\u4f8b\u5982\uff0c\u6709\u7684\u6a21\u578b\u76ee\u6807\u662f\u91cd\u5efa\u5bc6\u96c6\u7684\u56fe\u50cf\u50cf\u7d20\u6216\u662f\u91cd\u5efa\u8fde\u7eed\u7684\u5c40\u90e8\u533a\u57df\u7279\u5f81\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u503e\u5411\u4e8e\u5b66\u4e60\u5608\u6742\u7684\u7ec6\u8282\uff0c\u800c\u4e0d\u662f\u9ad8\u5c42\u7ed3\u6784\uff08\u5982\u6587\u6863\u5e03\u5c40\uff09\u3002\u5e76\u4e14\uff0c\u56fe\u50cf\u548c\u6587\u672c\u76ee\u6807\u7684\u7c92\u5ea6\u4e0d\u540c\u8fdb\u4e00\u6b65\u589e\u52a0\u4e86\u8de8\u6a21\u6001\u5bf9\u9f50\u5b66\u4e60\u7684\u96be\u5ea6\uff0c\u800c\u8de8\u6a21\u6001\u5bf9\u9f50\u5b66\u4e60\u5bf9\u591a\u6a21\u6001\u8868\u793a\u5b66\u4e60\u975e\u5e38\u5173\u952e\u3002 \u4e3a\u4e86\u514b\u670d\u6587\u672c\u548c\u56fe\u50cf\u5728\u9884\u8bad\u7ec3\u76ee\u6807\u4e0a\u7684\u5dee\u5f02\uff0c\u4fc3\u8fdb\u591a\u6a21\u6001\u8868\u5f81\u5b66\u4e60\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 LayoutLMv3\uff0c\u4ee5\u7edf\u4e00\u7684\u6587\u672c\u548c\u56fe\u50cf\u63a9\u7801\u5efa\u6a21\u76ee\u6807\u6765\u9884\u8bad\u7ec3\u591a\u6a21\u6001\u6a21\u578b\uff0c\u5373 LayoutLMv3 \u5b66\u4e60\u91cd\u5efa\u8bed\u8a00\u6a21\u6001\u7684\u906e\u76d6\u8bcd ID\uff0c\u5e76\u5bf9\u79f0\u5730\u91cd\u5efa\u56fe\u50cf\u6a21\u6001\u7684\u906e\u76d6\u56fe\u50cf\u5757 ID\u3002 \u6b64\u5916\uff0c\u5bf9\u4e8e\u6587\u6863\u6765\u8bf4\uff0c\u6bcf\u4e2a\u6587\u672c\u8bcd\u90fd\u5bf9\u5e94\u7740\u4e00\u4e2a\u56fe\u50cf\u5757\u3002\u4e3a\u4e86\u5b66\u4e60\u8fd9\u79cd\u8de8\u6a21\u6001\u7684\u5bf9\u9f50\u5173\u7cfb\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u63d0\u51fa\u4e86\u4e00\u4e2a\u8bcd\u5757\u5bf9\u9f50\u9884\u8bad\u7ec3\u76ee\u6807\uff0c\u901a\u8fc7\u9884\u6d4b\u4e00\u4e2a\u6587\u672c\u8bcd\u7684\u5bf9\u5e94\u56fe\u50cf\u5757\u662f\u5426\u88ab\u906e\u76d6\uff0c\u6765\u5b66\u4e60\u8fd9\u79cd\u8bed\u8a00\u2014\u2014\u56fe\u50cf\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u5173\u7cfb\u3002 \u5728\u6a21\u578b\u67b6\u6784\u8bbe\u8ba1\u4e0a\uff0cLayoutLMv3 \u4e0d\u4f9d\u8d56\u590d\u6742\u7684 CNN \u6216 Faster R-CNN \u7f51\u7edc\u6765\u8868\u5f81\u56fe\u50cf\uff0c\u800c\u662f\u76f4\u63a5\u5229\u7528\u6587\u6863\u56fe\u50cf\u7684\u56fe\u50cf\u5757\uff0c\u4ece\u800c\u5927\u5927\u8282\u7701\u4e86\u53c2\u6570\u5e76\u907f\u514d\u4e86\u590d\u6742\u7684\u6587\u6863\u9884\u5904\u7406\uff08\u5982\u4eba\u5de5\u6807\u6ce8\u76ee\u6807\u533a\u57df\u6846\u548c\u6587\u6863\u76ee\u6807\u68c0\u6d4b\uff09\u3002\u7b80\u5355\u7684\u7edf\u4e00\u67b6\u6784\u548c\u8bad\u7ec3\u76ee\u6807\u4f7f LayoutLMv3 \u6210\u4e3a\u901a\u7528\u7684\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u53ef\u9002\u7528\u4e8e\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u548c\u4ee5\u56fe\u50cf\u4e3a\u4e2d\u5fc3\u7684\u6587\u6863 AI \u4efb\u52a1\u3002 \u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728\u4e94\u4e2a\u6570\u636e\u96c6\u4e2d\u8bc4\u4f30\u4e86\u9884\u8bad\u7ec3\u7684 LayoutLMv3 \u6a21\u578b\uff0c\u5305\u62ec\u4ee5\u6587\u672c\u4e3a\u4e2d\u5fc3\u7684\u6570\u636e\u96c6\uff1a\u8868\u5355\u7406\u89e3 FUNSD \u6570\u636e\u96c6\uff0c\u7968\u636e\u7406\u89e3 [&hellip;]<\/p>\n","protected":false},"author":34512,"featured_media":1070820,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1069830","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069830","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":2,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069830\/revisions"}],"predecessor-version":[{"id":1070844,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069830\/revisions\/1070844"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1070820"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1069830"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1069830"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1069830"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1069830"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}