
{"id":12092,"date":"2026-09-10T20:02:19","date_gmt":"2026-09-10T12:02:19","guid":{"rendered":"https:\/\/infernews.com\/blog\/marigold-v2-revisiting-diffusion-transformers-for-monocular-depth-estimation\/"},"modified":"2026-09-10T20:02:19","modified_gmt":"2026-09-10T12:02:19","slug":"marigold-v2-revisiting-diffusion-transformers-for-monocular-depth-estimation","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/marigold-v2-revisiting-diffusion-transformers-for-monocular-depth-estimation\/","title":{"rendered":"Marigold V2\uff1a\u628a\u64f4\u6563 Transformer \u6539\u6210\u4e00\u6b65\u6df1\u5ea6\u4f30\u8a08\u6a21\u578b"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/teaser-1.jpg\" alt=\"Project website\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u904e\u53bb\u505a\u55ae\u76ee\u6df1\u5ea6\u4f30\u8a08\uff0c\u8981\u9ebc\u9760\u50b3\u7d71\u76e3\u7763\u5b78\u7fd2\uff0c\u8981\u9ebc\u7528\u64f4\u6563\u6a21\u578b\u758a\u597d\u591a\u6b65\u53bb\u566a\uff0c\u5169\u8005\u90fd\u5514\u6613\u517c\u9867\u7d30\u7bc0\u540c\u6548\u7387\u3002\u83ef\u70ba Bayer Lab \u958b\u6e90 Marigold V2 \u8d70\u7b2c\u4e09\u689d\u8def\uff1a\u76f4\u63a5\u628a\u9810\u8a13\u7df4\u5605\u64f4\u6563 Transformer\uff08DiT\uff09\u7576\u6210\u55ae\u6b65\u5bc6\u96c6\u9810\u6e2c\u5668\uff0c\u91cd\u65b0\u5fae\u8abf\u4e4b\u5f8c\u8f38\u51fa\u6df1\u5ea6\u3001\u900f\u8996\u6df1\u5ea6\u3001\u6cd5\u7dda\u3001\u53cd\u7167\u7387\u7b49\u591a\u7a2e\u5bc6\u96c6\u6a21\u614b\uff0c\u6bcb\u9808\u53cd\u8986\u53bb\u566a\u3002\u5c0d\u5f9e\u696d\u4eba\u54e1\u569f\u8b1b\uff0c\u5462\u500b\u6539\u52d5\u610f\u5473\u4f4f\u4e00\u5f35\u666e\u901a RGB \u76f8\u53ef\u4ee5\u5373\u6642\u8b8a\u6210\u5e36\u7d30\u7bc0\u5605\u6df1\u5ea6\u5716\uff0c\u6bcb\u9808\u7b49\u5e7e\u79d2\u53bb\u758a\u6b65\u9a5f\uff0c\u6bdb\u9aee\u908a\u7de3\u540c\u9aee\u7d72\u5462\u985e\u5e7c\u7bc0\u90fd\u4fdd\u7559\u5f97\u5230\uff0c\u8cbc\u8fd1 SOTA \u6c34\u6e96\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u6a21\u578b\u5bb6\u65cf\u540c\u8a13\u7df4\u6210\u672c<\/strong>\uff1a\u6bcf\u500b checkpoint \u5305\u542b\u4e00\u7d44 rank-128 LoRA \u9069\u914d\u5668\uff0c\u914d\u5408 4-bit \u91cf\u5316\u5605 DiT \u4e3b\u9ad4\uff0c\u500b\u5225\u4efb\u52d9\u4ef2\u6703\u9644\u5e36\u5fae\u8abf\u904e\u5605 VAE \u89e3\u78bc\u5668\u3002\u5e95\u5c64\u51cd\u7d50\u5605\u57fa\u790e\u6a21\u578b\u6703\u53e6\u5916\u4e0b\u8f09\u3002\u6574\u500b\u5fae\u8abf\u6d41\u7a0b\u7528\u55ae\u5f35\u6d88\u8cbb\u7d1a GPU \u5c11\u65bc\u4e00\u661f\u671f\u5c31\u5b8c\u6210\u5230\uff0c\u5c0d\u500b\u4eba\u958b\u767c\u8005\u540c\u7d30\u578b\u5be6\u9a57\u5ba4\u90fd\u4fc2\u53ef\u89f8\u53ca\u5605\u7bc4\u570d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7a0b\u5f0f\u78bc\u7d50\u69cb\u6e05\u6670\uff0c\u65b9\u4fbf\u4e8c\u6b21\u958b\u767c<\/strong>\uff1a\u5009\u5eab\u7528 manifest graph \u52a0 transform list \u5605\u65b9\u5f0f\u5b9a\u7fa9\u6578\u64da\u96c6\uff0c\u7db2\u7d61\u7d44\u4ef6\u540c\u640d\u5931\u51fd\u6578\u90fd\u4fc2\u8b80\u5beb batch dict \u5605\u6709\u5e8f\u6b65\u9a5f\uff0c\u6578\u64da\u96c6\u5b9a\u7fa9\u3001\u7db2\u7d61\u5716\u3001\u640d\u5931\u5716\u3001\u512a\u5316\u6392\u7a0b\u56db\u500b\u5c64\u6b21\u5206\u5f97\u5e7e\u958b\u3002\u5c0d\u60f3\u64f4\u5c55\u5230\u65b0\u4efb\u52d9\u5605\u4eba\u569f\u8b1b\uff0c\u53ea\u8981\u8ddf\u4f4f\u5462\u500b\u7d50\u69cb\u52a0\u6b65\u9a5f\u5c31\u5f97\uff0c\u5514\u4f7f\u7531\u96f6\u780c\u8d77\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5c0d\u6bd4\u50b3\u7d71\u591a\u6b65\u64f4\u6563\u6df1\u5ea6\u4f30\u8a08\uff0cMarigold V2 \u72a7\u7272\u5497\u8fed\u4ee3\u5f0f\u53bb\u566a\u5605\u9748\u6d3b\u6027\uff0c\u4f46\u63db\u569f\u4e00\u6b65\u63a8\u7406\u5605\u901f\u5ea6\u540c\u7d30\u7bc0\u9084\u539f\u5ea6\uff1b\u5c0d\u6bd4\u7d14\u76e3\u7763\u65b9\u6cd5\uff0c\u4f62\u501f\u7528\u5497\u64f4\u6563\u6a21\u578b\u5c0d\u7d0b\u7406\u540c\u908a\u754c\u5605\u5148\u5929\u7406\u89e3\u529b\uff0c\u518d\u52a0 LoRA \u4fdd\u6301\u8f15\u91cf\u3002\u540c\u4e00\u6279\u6b0a\u91cd\u4ef2\u53ef\u4ee5\u7528\u55ba metric depth completion \u7b49\u4e0b\u6e38\u4efb\u52d9\uff0c\u4e00\u500b\u6a21\u578b\u8986\u84cb\u591a\u500b\u5bc6\u96c6\u9810\u6e2c\u5834\u666f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">**\u9069\u5408\u505a AR\/VR\u30013D \u91cd\u5efa\u3001\u5f71\u50cf\u5408\u6210\u5605\u4e2d\u5c0f\u5718\u968a\uff0c\u6216\u8005\u60f3\u4f4e\u6210\u672c\u8a66\u9a57\u65b0\u5bc6\u96c6\u9810\u6e2c\u4efb\u52d9\u5605\u7814\u7a76\u8005\uff0c\u90fd\u6703\u89ba\u5f97\u5462\u500b setup \u5e7e\u5571\u624b\u3002HuggingFace \u4e0a\u6709\u5c0d\u61c9\u5605 demo \u7a7a\u9593\u53ef\u4ee5\u76f4\u63a5\u8a66\uff0c\u6b0a\u91cd\u540c\u8a13\u7df4\u63a8\u7406\u78bc\u90fd\u5df2\u958b\u6e90\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/huggingface.co\/spaces\/huawei-bayerlab\/marigold-v2-web\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/huawei-bayerlab\/marigold-v2\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/huawei-bayerlab\/marigold-v2-0\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u53ea\u8981\u4e00\u5f35\u666e\u901a RGB \u76f8\u7247\uff0c\u5c31\u5373\u6642\u8f38\u51fa\u9ad8\u8cea\u7d20\u6df1\u5ea6\u5716\uff0c\u9023\u6bdb\u9aee\u548c\u9aee\u7d72\u7b49\u5e7c\u7bc0\u90fd\u4fdd\u7559\u5230\u3002Marigold V2 \u628a\u9810\u8a13\u7df4\u64f4\u6563 Transformer \u6539\u9020\u70ba\u55ae\u6b65\u5bc6\u96c6\u9810\u6e2c\u6a21\u578b\u7684\u958b\u6e90\u9805\u76ee\uff0c\u4e00\u5f35\u6d88\u8cbb\u7d1a GPU \u4e00\u661f\u671f\u5167\u5c31\u80fd\u5b8c\u6210\u5fae\u8abf\u3002<\/p>\n","protected":false},"author":8,"featured_media":12091,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,182,76,127,185,30,132,150,199],"tags":[],"class_list":["post-12092","post","type-post","status-publish","format-standard","hentry","category-133","category-182","category-76","category-127","category-qwen","category-image","category-3d","category-150","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12092","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=12092"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12092\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/12091"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=12092"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=12092"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=12092"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}