{"id":6412,"date":"2025-10-21T10:18:21","date_gmt":"2025-10-21T02:18:21","guid":{"rendered":"https:\/\/nullthought.net\/?p=6412"},"modified":"2025-10-21T10:28:16","modified_gmt":"2025-10-21T02:28:16","slug":"deepseek-ocr%ef%bc%9a%e4%b8%8d%e6%98%af%e4%b8%80%e7%9b%ae%e5%8d%81%e8%a1%8c%ef%bc%8c%e8%80%8c%e6%98%af%e7%99%be%e8%a1%8c%e4%b9%83%e8%87%b3%e5%8d%83","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=6412","title":{"rendered":"DeepSeek-OCR\uff1a\u4e0d\u662f\u201c\u4e00\u76ee\u5341\u884c\u201d\uff0c\u800c\u662f\u201c\u767e\u884c\u201d\u4e43\u81f3\u201c\u5343\u884c\u201d\u7684\u4e0a\u4e0b\u6587\u7406\u89e3"},"content":{"rendered":"\n<p>\u5f53\u4eca\u5927\u6a21\u578b\u5728\u5904\u7406\u957f\u6587\u672c\u65f6\uff0c\u6ce8\u610f\u529b\u8ba1\u7b97\u968f\u957f\u5ea6\u4e8c\u6b21\u589e\u957f\uff0c\u7b97\u529b\u4e0e\u663e\u5b58\u538b\u529b\u5de8\u5927\u3002DeepSeek\u7684\u6700\u65b0\u6a21\u578b<strong><a href=\"https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-OCR\" target=\"_blank\" rel=\"noreferrer noopener\">DeepSeek-OCR<\/a><\/strong>\u63d0\u51fa\u7528\u201c\u89c6\u89c9\u2014\u6587\u672c\u201d\u901a\u9053\u8fdb\u884c\u201c\u5149\u5b66\u538b\u7f29\u201d\uff1a<strong>\u628a\u957f\u6587\u672c\u6e32\u67d3\u4e3a\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\uff0c\u7ecf\u89c6\u89c9\u7f16\u7801\u5668\u538b\u6210\u5c11\u91cf\u201c\u89c6\u89c9token\u201d\uff0c\u518d\u7531\u89e3\u7801\u5668\u8fd8\u539f\u4e3a\u6587\u672c<\/strong>\u3002\u4f5c\u8005\u4ee5\u7aef\u5230\u7aefOCR\u4e3a\u8bd5\u9a8c\u53f0\uff0c\u7cfb\u7edf\u540d\u4e3a DeepSeek-OCR\uff0c\u6838\u5fc3\u662f\u8f7b\u6fc0\u6d3b\u3001\u5c11\u89c6\u89c9token\u4e14\u80fd\u9ad8\u5206\u8fa8\u8f93\u5165\u7684 DeepEncoder\uff0c\u914d\u5408 DeepSeek-3B-MoE \u89e3\u7801\u5668\u3002\u5b9e\u8bc1\u8868\u660e\uff1a\u5728\u6587\u672ctoken\u4e0e\u89c6\u89c9token\u4e4b\u6bd4\u7ea6\u226410\u00d7\u65f6\uff0cOCR\u89e3\u7801\u7cbe\u5ea6\u53ef\u8fbe\u224897%\uff1b\u5373\u4f7f\u538b\u5230\u224820\u00d7\uff0c\u4ecd\u6709\u224860%\u7cbe\u5ea6\uff0c\u663e\u793a\u201c\u4ee5\u89c6\u52a9\u6587\u201d\u7684\u957f\u4e0a\u4e0b\u6587\u538b\u7f29\u8def\u7ebf\u53ef\u884c\u3002\u8be5\u4f53\u7cfb\u5728\u771f\u5b9e\u6587\u6863\u89e3\u6790\u57fa\u51c6 OmniDocBench \u4e0a\uff0c\u4f7f\u7528\u66f4\u5c11\u89c6\u89c9token\u53d6\u5f97\u4e0e\u6216\u4f18\u4e8eSOTA\u7684\u8868\u73b0\uff0c\u5e76\u5177\u5907\u56fe\u8868\u3001\u5316\u5b66\u5f0f\u3001\u51e0\u4f55\u56fe\u5f62\u7b49\u201c\u6df1\u89e3\u6790\u201d\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aHaoranWei, Yaofeng Sun, Yukun Li\uff0c\u6765\u81eaDeepSeek\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"454\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/10\/image-12-1024x454.png\" alt=\"\" class=\"wp-image-6413\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/10\/image-12-1024x454.png 1024w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/10\/image-12-300x133.png 300w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/10\/image-12-768x341.png 768w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/10\/image-12.png 1125w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">DeepSeek\u7684\u6700\u65b0\u6a21\u578b<strong><a href=\"https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-OCR\" target=\"_blank\" rel=\"noreferrer noopener\">DeepSeek-OCR<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<p>\u4e00\u3001\u6574\u4f53\u67b6\u6784\u4e0e\u6570\u636e\u6d41<br>\u7cfb\u7edf\u662f\u5178\u578b\u7684\u201c\u89c6\u89c9\u7f16\u7801\u5668 + \u8bed\u8a00\u89e3\u7801\u5668\u201d\u4e24\u6bb5\u5f0f\uff1a<br>1\uff09DeepEncoder\uff1a\u4e32\u8054\u4e24\u79cd\u6ce8\u610f\u529b\u98ce\u683c\u7684\u89c6\u89c9\u4e3b\u5e72\uff08\u524d\u534a\u4e3a\u4ee5\u7a97\u53e3\u6ce8\u610f\u529b\u4e3a\u4e3b\u7684 SAM-base\uff0c\u540e\u534a\u4e3a\u7a20\u5bc6\u5168\u5c40\u6ce8\u610f\u529b\u7684 CLIP-large\uff09\uff0c\u4e2d\u95f4\u63d2\u5165\u4e00\u4e2a\u201c16\u00d7\u5377\u79ef\u538b\u7f29\u5668\u201d\u628a\u89c6\u89c9token\u6570\u964d\u4e3a\u539f\u6765\u76841\/16\uff0c\u4ece\u800c\u5728\u4fdd\u6301\u9ad8\u5206\u8fa8\u8f93\u5165\u7684\u540c\u65f6\u63a7\u5236\u6fc0\u6d3b\u4e0etoken\u89c4\u6a21\u3002<br>2\uff09MoE \u89e3\u7801\u5668\uff1a\u91c7\u7528 DeepSeek-3B-MoE\uff0c\u63a8\u7406\u65f6\u6fc0\u6d3b\u7ea6570M\u53c2\u6570\uff086\/64\u8def\u7531\u4e13\u5bb6+2\u5171\u4eab\u4e13\u5bb6\uff09\uff0c\u4ee5\u5c0f\u6a21\u578b\u63a8\u7406\u6210\u672c\u83b7\u5f97\u63a5\u8fd13B\u8868\u8fbe\u529b\u3002<br>\u63a8\u7406\u65f6\uff0c\u56fe\u50cf\u2192DeepEncoder\uff08\u4ea7\u751f\u5c11\u91cf\u89c6\u89c9token\uff09\u2192MoE\u89e3\u7801\u5668\u6309\u63d0\u793a\u751f\u6210\u6587\u672c\u6216\u7ed3\u6784\u5316\u7ed3\u679c\u3002<\/p>\n\n\n\n<p>\u4e8c\u3001DeepEncoder \u7684\u5173\u952e\u8bbe\u8ba1<br>1\uff09\u53cc\u6bb5\u4e3b\u5e72\u7684\u52a8\u673a\uff1a\u7a97\u53e3\u6ce8\u610f\u529b\u6bb5\u627f\u62c5\u201c\u5c40\u90e8\/\u5bc6\u96c6\u611f\u77e5\u201d\uff0c\u6fc0\u6d3b\u5f00\u9500\u5c0f\uff1b\u7ecf16\u00d7\u5377\u79ef\u538b\u7f29\u540e\uff0c\u518d\u4ea4\u7ed9\u5168\u5c40\u6ce8\u610f\u529b\u6bb5\uff08CLIP-large\uff09\u505a\u201c\u5168\u5c40\/\u77e5\u8bc6\u6574\u5408\u201d\u3002\u75281024\u00d71024\u8f93\u5165\u793a\u4f8b\uff1a\u521d\u59cbpatch\u4e3a4096\u4e2atoken\uff0c\u7ecf\u538b\u7f29\u4e3a256\u4e2a\uff0c\u663e\u8457\u964d\u4f4e\u4e0b\u6e38\u6fc0\u6d3b\u4e0e\u5e8f\u5217\u957f\u5ea6\u3002<br>2\uff0916\u00d7\u5377\u79ef\u538b\u7f29\u5668\uff1a\u4e24\u5c42\u5377\u79ef\uff08kernel=3, stride=2, padding=1\uff09\uff0c\u901a\u9053\u4ece256\u21921024\uff1b\u5728\u4e0d\u727a\u7272\u8fc7\u591a\u53ef\u8fd8\u539f\u6027\u7684\u524d\u63d0\u4e0b\uff0c\u76f4\u63a5\u628atoken\u6570\u7f2916\u500d\u3002<br>3\uff09\u591a\u5206\u8fa8\u4e0e\u591a\u6a21\u5f0f\uff1a\u539f\u751f Tiny\/Small\/Base\/Large \u5206\u522b\u5bf9\u5e94512\u00b2\/640\u00b2\/1024\u00b2\/1280\u00b2\u8f93\u5165\uff0c\u8f93\u51fa\u89c6\u89c9token\u4e3a64\/100\/256\/400\uff1b\u4e3a\u517c\u987e\u8d85\u9ad8\u5206\u8fa8\u6587\u6863\uff0c\u53c8\u5f15\u5165\u201cGundam\u201d\u52a8\u6001\u5206\u8fa8\u7387\uff08n\u5757\u5c40\u90e8tile+1\u5e45\u5168\u5c40\u89c6\u56fe\uff0cn\u2208[2,9]\uff09\uff0c\u8f93\u51fatoken\u4e3a n\u00d7100+256\uff08\u6216\u66f4\u5927Master\u6a21\u5f0f\uff09\u3002Base\/Large\u91c7\u7528padding\u4fdd\u6301\u7eb5\u6a2a\u6bd4\uff0c\u5e76\u7ed9\u51fa\u201c\u6709\u6548\u89c6\u89c9token\u201d\u8ba1\u7b97\u516c\u5f0f\u4ee5\u53cd\u6620padding\u540e\u7684\u6709\u6548\u4fe1\u606f\u5bc6\u5ea6\u3002<\/p>\n\n\n\n<p>\u4e09\u3001\u538b\u7f29\u2014\u8fd8\u539f\u80fd\u529b\u7684\u5b9a\u91cf\u7ed3\u679c\u4e0e\u8fb9\u754c<br>\u4f5c\u8005\u5728 Fox \u6587\u6863\u57fa\u51c6\u7684\u82f1\u6587600\u20131300\u6587\u672ctoken\u9875\u9762\u4e0a\u7cfb\u7edf\u6d4b\u8bc4\uff08100\u9875\uff09\uff1a<br>\u2022 \u5f53\u89c6\u89c9token=64\uff08Tiny\uff0c512\u00b2\uff09\u65f6\uff0c\u6587\u672c\/\u89c6\u89c9\u224810.5\u00d7\u65f6\u7cbe\u5ea6\u224896.5%\uff1b\u224815\u201319.7\u00d7\u65f6\u7cbe\u5ea6\u964d\u81f3\u224859\u201386%\u3002<br>\u2022 \u5f53\u89c6\u89c9token=100\uff08Small\uff0c640\u00b2\uff09\u65f6\uff0c\u22486.7\u201312.6\u00d7\u533a\u95f4\u7cbe\u5ea6\u224887\u201398.5%\u3002<br>\u7ed3\u8bba\uff1a\u224810\u00d7\u9644\u8fd1\u53ef\u8fbe\u201c\u8fd1\u65e0\u635f\u201dOCR\u89e3\u7801\uff1b\u8d85\u8fc710\u00d7\u540e\u5f00\u59cb\u201c\u53ef\u63a7\u9000\u5316\u201d\uff0c\u5230\u224820\u00d7\u4ecd\u80fd\u4fdd\u6301\u224860%\u6587\u672c\u8fd1\u4f3c\u53ef\u8fd8\u539f\u3002\u8fd9\u4e3a\u201c\u5386\u53f2\u5bf9\u8bdd\u8f6c\u56fe\u50cf\u2192\u591a\u7ea7\u964d\u91c7\u6837\u201d\u7684\u8bb0\u5fc6\u8870\u9000\/\u9057\u5fd8\u673a\u5236\u63d0\u4f9b\u4e86\u7ecf\u9a8c\u4e0a\u9650\u4e0e\u8bbe\u8ba1\u6293\u624b\u3002<\/p>\n\n\n\n<p>\u56db\u3001\u5728 OmniDocBench \u4e0a\u7684\u5b9e\u7528\u8868\u73b0<br>OmniDocBench \u7efc\u5408\u8bc4\u4f30\u201c\u6574\u4f53\/\u7eaf\u6587\u672c\/\u516c\u5f0f\/\u8868\u683c\/\u987a\u5e8f\u201d\u7b49\u7ef4\u5ea6\u7684\u7f16\u8f91\u8ddd\u79bb\uff08\u8d8a\u5c0f\u8d8a\u597d\uff09\u3002DeepSeek-OCR \u5728\u89c6\u89c9token\u66f4\u5c11\u7684\u524d\u63d0\u4e0b\u8fbe\u5230\u6216\u903c\u8fd1SOTA\uff1a<br>\u2022 Small\uff08100 token\uff09\u5df2\u663e\u8457\u4f18\u4e8e\u82e5\u5e72\u7aef\u5230\u7aef\u6a21\u578b\uff1b<br>\u2022 Base\uff08256\uff1b\u6709\u6548\u2248182\uff09\/Large\uff08400\uff1b\u6709\u6548\u2248285\uff09\u903c\u8fd1\u6216\u5bf9\u6807\u66f4\u5927\u5f00\u9500\u6a21\u578b\uff1b<br>\u2022 Gundam\uff08&lt;800 token\uff09\u6574\u4f53\u4f18\u4e8e MinerU2.0\uff08\u22486790 token\uff09\uff1b<br>\u2022 \u628a\u539f\u56fe\u7edf\u4e00\u5230200dpi\u7684 Gundam-M \u53ef\u8fdb\u4e00\u6b65\u964d\u8bef\u5dee\u3002<br>\u6b64\u5916\uff0c\u4e0d\u540c\u6587\u6863\u7c7b\u578b\u5bf9token\u9884\u7b97\u6781\u4e0d\u5747\u5300\uff1a\u5e7b\u706f\u7247\u53ea\u970064 token\u5373\u6709\u5f88\u597d\u6548\u679c\uff1b\u4e66\u7c4d\/\u62a5\u544a\u5728100 token\u8868\u73b0\u4f18\u826f\uff1b\u62a5\u7eb8\u56e0\u6bcf\u9875\u6587\u672ctoken\u5e38\u8fbe4\u20135k\uff0c\u9700\u8981 Gundam \u6216 Gundam-M\u3002\u5b9e\u8df5\u4e0a\u636e\u6b64\u53ef\u6309\u201c\u9875\u5185\u6587\u672c\u5bc6\u5ea6\u201d\u4e0e\u201c\u7248\u5f0f\u590d\u6742\u5ea6\u201d\u81ea\u9002\u5e94\u5206\u914d\u89c6\u89c9token\u9884\u7b97\u3002<\/p>\n\n\n\n<p>\u4e94\u3001\u201c\u6df1\u89e3\u6790\u201d\u4e0e\u591a\u4efb\u52a1\u80fd\u529b<br>\u9664\u5e38\u89c4\u201c\u7248\u9762+\u6587\u5b57\u201d\u5916\uff0c\u6a21\u578b\u8fd8\u80fd\u4ee5\u7edf\u4e00\u63d0\u793a\u5bf9\u9875\u5185\u5b50\u56fe\u8fdb\u884c\u4e8c\u6b21\u89e3\u6790\uff1a<br>\u2022 \u56fe\u8868\uff1a\u8f6c\u4e3a\u7ed3\u6784\u5316HTML\u8868\u683c\uff1b<br>\u2022 \u5316\u5b66\u5f0f\uff1a\u6e32\u67d3\u8bc6\u522b\u2192SMILES \u8fd8\u539f\uff1b<br>\u2022 \u51e0\u4f55\u56fe\uff1a\u4ee5\u201c\u6162\u611f\u77e5\u201d\u5f0f\u5b57\u5178\u7ed3\u6784\u8f93\u51fa\u7ebf\u6bb5\u3001\u7aef\u70b9\u4e0e\u7c7b\u578b\uff1b<br>\u2022 \u81ea\u7136\u56fe\u50cf\uff1a\u81f4\u5bc6\u63cf\u8ff0\u3001\u76ee\u6807\u5b9a\u4f4d\u3001\u68c0\u6d4b\u4e0eOCR\u6df7\u5408\u4efb\u52a1\u3002<br>\u8fd9\u4e9b\u80fd\u529b\u6765\u81ea\u201cOCR 2.0\u201d\u4e0e\u201c\u901a\u7528\u89c6\u89c9\u6570\u636e\u201d\u7684\u8054\u5408\u8bad\u7ec3\uff0c\u867d\u975e\u901a\u7528VLM\uff0c\u4f46\u4fdd\u7559\u4e86\u5fc5\u8981\u7684\u89c6\u89c9\u63a5\u53e3\uff0c\u65b9\u4fbf\u6269\u5c55\u3002<\/p>\n\n\n\n<p>\u516d\u3001\u6570\u636e\u5f15\u64ce\u4e0e\u914d\u6bd4<br>\u8bad\u7ec3\u6570\u636e\u5360\u6bd4\uff1aOCR\u224870%\u3001\u901a\u7528\u89c6\u89c9\u224820%\u3001\u7eaf\u6587\u672c\u224810%\uff0c\u5e8f\u5217\u957f\u5ea6\u7edf\u4e00\u52308192\u3002<br>\u2022 OCR 1.0\uff1a\u591a\u8bedPDF\uff08\u224830M\u9875\uff0c\u542b\u22482M\u4e2d\u82f1\u7cbe\u6807\u5e03\u5c40+\u8bc6\u522b\u4ea4\u9519\u6807\u6ce8\uff09\uff0c\u573a\u666f\u6587\u5b57\uff08\u4e2d\u82f1\u5404\u224810M\uff0cPaddleOCR\u81ea\u52a8\u6807\u6ce8\uff09\u3002<br>\u2022 OCR 2.0\uff1a\u56fe\u8868\uff08\u224810M\uff0cmatplotlib\/pyecharts\u6e32\u67d3\u2192HTML\u8868\uff09\u3001\u5316\u5b66\u5f0f\uff08\u22485M\uff0cSMILES\u2192RDKit\u6e32\u67d3\uff09\u3001\u51e0\u4f55\uff08\u22481M\uff0c\u201c\u7ffb\u8bd1\u4e0d\u53d8\u201d\u589e\u5f3a+\u6162\u611f\u77e5\u7f16\u7801\uff09\u3002<br>\u8fd9\u5957\u201c\u7c97\u6807+\u7cbe\u6807+\u98de\u8f6e\u81ea\u4e3e\u201d\u7684\u6570\u636e\u5de5\u7a0b\uff0c\u662f\u7aef\u5230\u7aefOCR\u53ef\u843d\u5730\u4e0e\u53ef\u6269\u5c55\u7684\u5173\u952e\u3002<\/p>\n\n\n\n<p>\u4e03\u3001\u8bad\u7ec3\u6d41\u6c34\u7ebf\u4e0e\u5de5\u7a0b\u5b9e\u73b0<br>\u4e24\u9636\u6bb5\u8bad\u7ec3\uff1a<br>1\uff09\u5148\u8bad DeepEncoder\uff1a\u4ee5\u7d27\u51d1LM\u505a\u201c\u4e0b\u4e00token\u9884\u6d4b\u201d\uff0c\u5728OCR 1.0\/2.0\u4e0eLAION\u91c7\u6837\u6570\u636e\u4e0a\u8bad\u7ec3\uff0c2\u4e2aepoch\uff0cbatch=1280\uff0cAdamW+\u4f59\u5f26\u9000\u706b\uff0clr=5e-5\uff0c\u5e8f\u5217\u957f4096\u3002<br>2\uff09\u518d\u8bad\u6574\u6a21 DeepSeek-OCR\uff1a\u7528PP=4\u7684\u6d41\u6c34\u7ebf\u5e76\u884c\uff08SAM+\u538b\u7f29\u5668\u5728PP0\u5e76\u51bb\u7ed3\uff1bCLIP\u5728PP1\u53ef\u8bad\uff1bMoE\u768412\u5c42\u5728PP2\/PP3\u54046\u5c42\uff09\uff0c20\u8282\u70b9\u00d78\u00d7A100-40G\uff0cDP=40\uff0c\u5168\u5c40batch=640\uff1b\u591a\u6a21\u6001\u65e5\u541e\u5410\u224870B token\uff0c\u6587\u672c\u224890B\u3002Gundam-M \u901a\u8fc7\u5728\u5df2\u8bad\u6a21\u578b\u4e0a\u7ee7\u7eed\u8bad\u7ec3\u5f97\u5230\u3002<\/p>\n\n\n\n<p>\u516b\u3001\u4e0e\u5178\u578b\u89c6\u89c9\u7f16\u7801\u65b9\u6848\u7684\u5bf9\u6bd4<br>\u2022 \u53cc\u5854\u5e76\u884c\uff08\u5982Vary\u7cfb\uff09\uff1a\u9700\u8981\u53cc\u91cd\u9884\u5904\u7406\u3001\u8bad\u7ec3\u5e76\u884c\u590d\u6742\u3002<br>\u2022 \u5e73\u94fa\u5207\u5757\uff08\u5982InternVL2\u7cfb\uff09\uff1a\u80fd\u4e0a\u6781\u9ad8\u5206\u8fa8\u7387\uff0c\u4f46\u539f\u751f\u5206\u8fa8\u7387\u4f4e\u6613\u788e\u7247\u5316\u3001token\u6570\u81a8\u80c0\u3002<br>\u2022 \u81ea\u9002\u5e94\u5206\u8fa8\u7387\uff08\u5982Qwen-VL\/NaViT\uff09\uff1a\u7075\u6d3b\u4f46\u5927\u56fe\u6fc0\u6d3b\u5185\u5b58\u6781\u9ad8\uff0c\u5e8f\u5217\u6253\u5305\u4e5f\u62c9\u957f\u8bad\u7ec3\u5e8f\u5217\u3002<br>DeepEncoder \u4ee5\u201c\u7a97\u53e3\u6ce8\u610f\u529b\u219216\u00d7\u538b\u7f29\u2192\u5168\u5c40\u6ce8\u610f\u529b\u201d\u7684\u4e32\u8054\u5f0f\u6298\u4e2d\uff0c\u65e2\u4fdd\u7559\u9ad8\u5206\u8fa8\uff0c\u53c8\u628a\u6fc0\u6d3b\u4e0etoken\u89c4\u6a21\u538b\u5230\u201c\u7ed3\u6784\u5408\u7406\u201d\u7684\u8303\u7574\u3002<\/p>\n\n\n\n<p>\u4e5d\u3001\u7406\u8bba\u542f\u53d1\uff1a\u4ece\u201c\u5149\u5b66\u538b\u7f29\u201d\u5230\u201c\u8bb0\u5fc6\u9057\u5fd8\u201d<br>\u4f5c\u8005\u63d0\u51fa\u628a\u5bf9\u8bdd\u5386\u53f2\u6e32\u67d3\u6210\u56fe\u50cf\u5e76\u968f\u65f6\u95f4\u4e0b\u91c7\u6837\uff1a\u8fd1\u671f\u9ad8\u5206\u8fa8\u3001\u8fdc\u671f\u4f4e\u5206\u8fa8\uff0c\u914d\u5408\u89c6\u89c9token\u9884\u7b97\u9012\u51cf\uff0c\u6a21\u62df\u4eba\u7c7b\u8bb0\u5fc6\u8870\u9000\u66f2\u7ebf\uff0c\u5f62\u6210\u201c\u6210\u672c\u53ef\u63a7\u3001\u4fe1\u606f\u968f\u65f6\u8870\u51cf\u201d\u7684\u65e0\u9650\u4e0a\u4e0b\u6587\u6846\u67b6\u96cf\u5f62\u3002\u5f53\u524d\u5728\u224810\u00d7\u9644\u8fd1\u63a5\u8fd1\u201c\u65e0\u635f\u201d\uff0c&gt;10\u00d7\u5448\u201c\u4f18\u96c5\u9000\u5316\u201d\uff0c\u4e3a\u5206\u5c42\u7f13\u5b58\/\u56de\u653e\u63d0\u4f9b\u786c\u6307\u6807\u3002<\/p>\n\n\n\n<p>\u5341\u3001\u5c40\u9650\u6027\u4e0e\u98ce\u9669\u70b9<br>1\uff09\u7248\u5f0f\/\u5b57\u4f53\u6781\u590d\u6742\u3001\u957f\u9875\u8d85\u5bc6\u65f6\u4ecd\u9700\u8f83\u9ad8token\uff08Gundam\u5bb6\u65cf\uff09\uff0c\u541e\u5410\u4f1a\u53d7\u9650\u3002<br>2\uff09\u56fe\u50cf\u6e32\u67d3\u4e0e\u538b\u7f29\u94fe\u8def\u4e2d\u7684\u201c\u5fae\u6a21\u7cca\/\u952f\u9f7f\/\u63d2\u503c\u8bef\u5dee\u201d\u4f1a\u7d2f\u79ef\u5230\u6587\u672c\u8bef\u5dee\uff0c\u5c24\u5176&gt;10\u00d7\u533a\u95f4\u3002<br>3\uff09\u7f3a\u5c11\u4e13\u95e8\u7684\u5bf9\u8bdd\u7ea7\u9488\u5806\u68c0\u7d22\uff08NIH\uff09\u4e0e\u6570\u5b57\u2014\u5149\u5b66\u4ea4\u66ff\u9884\u8bad\u7ec3\u9a8c\u8bc1\uff1b\u8de8\u57df\u6cdb\u5316\u5230\u975e\u6587\u6863\u77e5\u8bc6\u68c0\u7d22\u573a\u666f\u4ecd\u9700\u989d\u5916\u8bc1\u636e\u3002<br>4\uff09MoE \u8def\u7531\u7a33\u5b9a\u6027\u4e0e\u90e8\u7f72\u590d\u6742\u5ea6\u5728\u8d85\u5927\u89c4\u6a21\u670d\u52a1\u4e0b\u9700\u66f4\u591a\u5de5\u7a0b\u5b9e\u8df5\u3002<\/p>\n\n\n\n<p>\u5341\u4e00\u3001\u5de5\u7a0b\u843d\u5730\u5efa\u8bae\uff08\u7ed9\u7814\u53d1\u4e0e\u4ea7\u54c1\u56e2\u961f\uff09<br>1\uff09\u81ea\u9002\u5e94token\u9884\u7b97\uff1a\u4e0a\u7ebf\u65f6\u4ee5\u201c\u9875\u5185\u6587\u672c\u5bc6\u5ea6\u4f30\u8ba1+\u7248\u5f0f\u7c7b\u522b\u8bc6\u522b\u201d\u8def\u7531\u5230 Tiny\/Small\/Base\/Large\/Gundam\uff1b\u62a5\u7eb8\u3001\u516c\u544a\u3001\u5e74\u62a5\u7b49\u9ad8\u5bc6\u6587\u6863\u4f18\u5148Gundam\u3002<br>2\uff09\u63d0\u793a\u6a21\u677f\u5e93\uff1a\u7edf\u4e00\u201cFree OCR\u201d\u201c\u5e26\u7248\u9762Markdown\u201d\u201cParse the figure\u201d\u7b49\u6a21\u5f0f\uff0c\u6309\u4efb\u52a1\u81ea\u52a8\u5207\u6362\u8f93\u51fa\u534f\u8bae\uff08\u7eaf\u6587\u3001HTML\u8868\u3001SMILES\u3001\u51e0\u4f55\u5b57\u5178\uff09\u3002<br>3\uff09\u6570\u636e\u98de\u8f6e\uff1a\u628a\u7ebf\u4e0a\u89e3\u6790\u7ed3\u679c\u7684\u4eba\u5ba1\u7ea0\u9519\u56de\u704c\u5230\u7ec6\u6807\u96c6\uff1b\u56fe\u8868\/\u51e0\u4f55\/\u5316\u5b66\u5f0f\u6301\u7eed\u5408\u6210\u589e\u5f3a\uff0c\u63a7\u4f4f\u957f\u5c3e\u3002<br>4\uff09\u8bad\u7ec3\/\u63a8\u7406\u7f16\u6392\uff1a\u4fdd\u6301SAM+\u538b\u7f29\u5668\u51bb\u7ed3\uff0cCLIP\u7aef\u4e0eMoE\u7aef\u6309\u65b0\u57df\u5c0f\u6b65\u5fae\u8c03\uff1b\u670d\u52a1\u4fa7\u4ee5PP+TP\u6df7\u5408\u7f16\u6392\uff0c\u89c6\u89c9token\u4e0a\u9650\u505a\u5f39\u6027\u914d\u989d\u3002<br>5\uff09\u8bc4\u6d4b\u4e0e\u56de\u5f52\uff1a\u9664OmniDocBench\u5916\uff0c\u8865\u5145\u6309\u4e1a\u52a1\u573a\u666f\u6784\u9020\u7684\u9875\u7ea7E2E\u6307\u6807\uff08\u7ed3\u6784\u8fd8\u539fF1\u3001\u5b57\u6bb5\u53ec\u56de\u3001\u8868\u683c\u5355\u5143\u683cED\uff09\uff0c\u5e76\u968f\u5206\u8fa8\u7387\/\u538b\u7f29\u6bd4\u7ed8\u5236\u201c\u4ee3\u4ef7\u2014\u8d28\u91cf\u201d\u66f2\u7ebf\u3002<\/p>\n\n\n\n<p>\u5341\u4e8c\u3001\u5173\u952e\u516c\u5f0f\u4e0e\u6307\u6807\u89e3\u8bfb<br>1\uff09\u6709\u6548\u89c6\u89c9token\uff1a\u5f53\u91c7\u7528padding\u4fdd\u6301\u7eb5\u6a2a\u6bd4\u65f6\uff0c\u6709\u6548token \u2248 \u2308\u5b9e\u9645token \u00d7 [1 \u2212 (max(w,h)\u2212min(w,h))\/max(w,h)]\u2309\uff0c\u7528\u4e8e\u4f30\u7b97\u771f\u6b63\u627f\u8f7d\u5185\u5bb9\u7684token\u6bd4\u4f8b\u3002<br>2\uff09\u538b\u7f29\u6bd4\u5b9a\u4e49\uff1a\u6587\u672ctoken\u6570 \/ \u89c6\u89c9token\u6570\u3002\u5efa\u8bae\u4ee5\u201c\u9875\u7ea7\u6587\u672ctoken\u4f30\u8ba1\u5668\u201d\u5728\u4e0a\u7ebf\u65f6\u5148\u4f30\uff0c\u518d\u7ed9\u51fa\u76ee\u6807\u89c6\u89c9token\u4e0a\u9650\uff0c\u529b\u4e89\u843d\u5728\u226410\u00d7\u5e26\u5bbd\u5185\u3002<br>3\uff09\u7f16\u8f91\u8ddd\u79bb\uff08ED\uff09\uff1a\u7528\u4e8eOmniDocBench\u6574\u4f53\u4e0e\u5b50\u4efb\u52a1\uff08\u6587\u672c\/\u516c\u5f0f\/\u8868\u683c\/\u987a\u5e8f\uff09\u8bc4\u4f30\uff0c\u8d8a\u5c0f\u8d8a\u597d\uff0c\u4fbf\u4e8e\u8de8\u6a21\u578b\u3001\u8de8token\u9884\u7b97\u7684\u516c\u5e73\u6bd4\u8f83\u3002<\/p>\n\n\n\n<p>\u5341\u4e09\u3001\u4e0e\u540c\u7c7b\u65b9\u6cd5\u7684\u91cf\u5316\u5bf9\u7167\u6458\u8981<br>\u2022 \u76f8\u5bf9 GOT-OCR2.0\uff08256 token\/\u9875\uff09\uff1aDeepSeek-OCR \u4ee5100 token\/\u9875\u5373\u80fd\u5728\u591a\u9879\u6307\u6807\u4e0a\u8d85\u8d8a\u3002<br>\u2022 \u76f8\u5bf9 MinerU2.0\uff08\u22486790 token\/\u9875\uff09\uff1aDeepSeek-OCR \u7684 Gundam\uff08&lt;800 token\/\u9875\uff09\u6574\u4f53\u66f4\u4f18\uff0c\u6027\u4ef7\u6bd4\u4f18\u52bf\u663e\u8457\u3002<br>\u2022 \u4e0e\u66f4\u5927\u53c2\u6570\u901a\u7528<strong><a href=\"https:\/\/nullthought.net\/?p=4920\" target=\"_blank\" rel=\"noreferrer noopener\">VLM<\/a><\/strong>\uff08InternVL\/Qwen-VL\/\u2026\uff09\u76f8\u6bd4\uff1a\u5728\u6587\u6863\u89e3\u6790\u8fd9\u4e00\u5bc6\u96c6\u6587\u672c\u4efb\u52a1\u4e0a\uff0c\u4e13\u95e8\u5316\u89c6\u89c9\u7f16\u7801\u4e0etoken\u538b\u7f29\u7b56\u7565\u5e26\u6765\u66f4\u597d\u7684\u201c\u8d28\u91cf\/\u6210\u672c\u201d\u5e73\u8861\u3002<\/p>\n\n\n\n<p>\u5341\u56db\u3001\u603b\u7ed3\u4e0e\u5c55\u671b<br>DeepSeek-OCR \u4ee5\u201c\u7a97\u53e3\u6ce8\u610f\u529b\u219216\u00d7\u5377\u79ef\u538b\u7f29\u2192\u5168\u5c40\u6ce8\u610f\u529b\u201d\u7684DeepEncoder\uff0c\u52a0\u4e0a\u9ad8\u6548MoE\u89e3\u7801\uff0c\u5c55\u793a\u4e86\u201c\u89c6\u89c9\u627f\u8f7d\u957f\u6587\u672c\u3001\u224810\u00d7\u8fd1\u65e0\u635f\u538b\u7f29\u201d\u7684\u53ef\u884c\u8def\u5f84\uff0c\u5e76\u5728\u5b9e\u9645\u6587\u6863\u4efb\u52a1\u4e2d\u4ee5\u66f4\u5c11token\u8fbe\u5230\u4e0e\u6216\u4f18\u4e8eSOTA\u7684\u6548\u679c\u3002\u540e\u7eed\u503c\u5f97\u63a8\u8fdb\uff1a\u6570\u5b57\u2014\u5149\u5b66\u4ea4\u66ff\u9884\u8bad\u7ec3\u3001\u9488\u5806\u68c0\u7d22\u573a\u666f\u9a8c\u8bc1\u3001\u8de8\u57df\u77e5\u8bc6\u62bd\u53d6\u3001\u4ee5\u53ca\u628a\u201c\u5206\u5c42\u5206\u8fa8\u7387+token\u914d\u989d\u201d\u843d\u5230\u5bf9\u8bdd\u7cfb\u7edf\u7684\u957f\u671f\u8bb0\u5fc6\u7ba1\u7406\u4e2d\uff0c\u5f62\u6210\u53ef\u89e3\u91ca\u3001\u53ef\u914d\u7f6e\u3001\u53ef\u5ea6\u91cf\u7684\u201c\u89c6\u89c9\u5316\u4e0a\u4e0b\u6587\u5185\u5b58\u201d\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>DeepSeek<a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-OCR\">&#8211;<\/a>OCR on GitHub: <a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-OCR\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/deepseek-ai\/DeepSeek-OCR<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5f53\u4eca\u5927\u6a21\u578b\u5728\u5904\u7406\u957f\u6587\u672c\u65f6\uff0c\u6ce8\u610f\u529b\u8ba1\u7b97\u968f\u957f\u5ea6\u4e8c\u6b21\u589e\u957f\uff0c\u7b97\u529b\u4e0e\u663e\u5b58\u538b\u529b\u5de8\u5927\u3002DeepSeek\u7684\u6700\u65b0\u6a21\u578bDeepSe [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,96,86,80],"class_list":["post-6412","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-deepseek","tag-llm","tag-cv"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u5f53\u4eca\u5927\u6a21\u578b\u5728\u5904\u7406\u957f\u6587\u672c\u65f6\uff0c\u6ce8\u610f\u529b\u8ba1\u7b97\u968f\u957f\u5ea6\u4e8c\u6b21\u589e\u957f\uff0c\u7b97\u529b\u4e0e\u663e\u5b58\u538b\u529b\u5de8\u5927\u3002DeepSeek\u7684\u6700\u65b0\u6a21\u578bDeepSe&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6412","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6412"}],"version-history":[{"count":2,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6412\/revisions"}],"predecessor-version":[{"id":6415,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6412\/revisions\/6415"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6412"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6412"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6412"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}