{"id":8196,"date":"2026-02-11T04:52:00","date_gmt":"2026-02-10T19:52:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8196"},"modified":"2026-02-01T10:59:15","modified_gmt":"2026-02-01T01:59:15","slug":"deepseek-ocr-2-visual-causal-flow","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8196","title":{"rendered":"DeepSeek-OCR 2: Visual Causal Flow\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>DeepSeek-OCR 2: Visual Causal Flow\u00a0<\/strong>[15.6]<br>\u672c\u7a3f\u3067\u306f,\u65b0\u3057\u3044\u30a8\u30f3\u30b3\u30fc\u30c0-\u30c7\u30a3\u30fc\u30d7\u30a8\u30f3\u30b3\u30fc\u30c0V2\u306e\u5b9f\u73fe\u53ef\u80fd\u6027\u3092\u691c\u8a0e\u3059\u308b\u305f\u3081\u306bDeepSeek-OCR 2\u3092\u63d0\u6848\u3059\u308b\u3002 DeepEncoder V2\u306f\u3001\u30a8\u30f3\u30b3\u30fc\u30c0\u306b\u56e0\u679c\u63a8\u8ad6\u6a5f\u80fd\u3092\u6301\u305f\u305b\u308b\u3088\u3046\u306b\u8a2d\u8a08\u3055\u308c\u3066\u304a\u308a\u3001\u30b3\u30f3\u30c6\u30f3\u30c4\u89e3\u91c8\u306e\u524d\u306b\u8996\u899a\u30c8\u30fc\u30af\u30f3\u3092\u30a4\u30f3\u30c6\u30ea\u30b8\u30a7\u30f3\u30c8\u306b\u4e26\u3079\u66ff\u3048\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3002 \u672c\u7814\u7a76\u306f,2\u6b21\u5143\u56e0\u679c\u63a8\u8ad6\u69cb\u9020\u3092\u7528\u3044\u30662\u6b21\u5143\u753b\u50cf\u7406\u89e3\u3092\u52b9\u679c\u7684\u306b\u5b9f\u73fe\u3067\u304d\u308b\u304b\u5426\u304b\u3068\u3044\u3046,\u65b0\u3057\u3044\u30d1\u30e9\u30c0\u30a4\u30e0\u3092\u63a2\u6c42\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2601.20552v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2601.20552v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 28 Jan 2026 12:46:07 GMT)<\/li>\n\n\n\n<li>DeepEncoder V2\u3068DeepSeek-OCR 2\u306e\u63d0\u6848\u3002\u5f37\u529b\u306a\u6027\u80fd\u3092\u9054\u6210\u3002\u7279\u306bDeepEncode V2\u306b\u306f\u300cDeepEncoder V2, featuring several key innovations: (1) we replace the CLIP [37] component in DeepEncoder [54] with a compact LLM [48] architecture, as illustrated in Figure 1, to achieve visual causal flow; (2) to enable parallelized processing, we introduce learnable queries [10], termed causal flow tokens, with visual tokens prepended as a prefix\u2014through a customized attention mask, visual tokens maintain global receptive fields, while causal flow tokens can obtain visual token reordering ability; (3) we maintain equal cardinality between causal and visual tokens (with redundancy such as padding and borders) to provide sufficient capacity for re-fixation; (4) only the causal flow tokens\u2014the latter half of the encoder outputs\u2014are fed to the LLM [24] decoder, enabling cascade causal-aware visual understanding.\u300d\u3068\u304b\u306a\u308a\u306e\u5909\u66f4\u304c\u306a\u3055\u308c\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/deepseek-ai\/DeepSeek-OCR-2\">GitHub &#8211; deepseek-ai\/DeepSeek-OCR-2: Visual Causal Flow<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[285],"class_list":["post-8196","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-ocr"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8196","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8196"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8196\/revisions"}],"predecessor-version":[{"id":8197,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8196\/revisions\/8197"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8196"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8196"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8196"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}