{
  "slug": "multimodal-representations",
  "category": "多模态学习",
  "title": "从图像分块到图文对齐：ViT、CLIP 与多模态表示的边界",
  "dek": "区分视觉骨干、跨模态训练目标和输入压缩，说明“图文相似”为什么不能直接等同于完整场景理解。",
  "english": "ViT, CLIP, and Perceiver address different representation bottlenecks. We distinguish tokenization, contrastive supervision, and latent compression with a transparent patch-count calculation.",
  "refs": [
    "vit",
    "clip",
    "perceiver"
  ],
  "sections": [
    [
      "摘要",
      "多模态模型并非简单把图片塞进语言模型。输入表示、训练目标与信息压缩方式，决定了什么信息被保留、怎样进行匹配，以及如何评估。本文以 ViT、CLIP 与 Perceiver 为线索，解释三类不同问题，并用图像分块的解析计算展示分辨率与注意力代价之间的关系。",
      [
        "vit",
        "clip",
        "perceiver"
      ]
    ],
    [
      "1. ViT 把图像组织为词元序列",
      "ViT 将图像划分为块，映射为向量，再通过 Transformer 处理。论文的重要条件包括大规模预训练与后续迁移，并不等于“任何小数据集上，纯 Transformer 都必然优于卷积”。架构比较应保留预训练数据、增强方式与计算预算；否则数据投入差异会被误写为结构的单独贡献。",
      [
        "vit"
      ]
    ],
    [
      "2. CLIP 用文本改变监督接口",
      "CLIP 在图文配对数据上学习对比表示，把匹配的图像与文本拉近、与不匹配组合区分。论文报告的训练数据规模为 4 亿图文对。推理时，可将候选类别写成文本，用相似度进行零样本分类。这里“零样本”指没有针对该下游分类训练集做相应训练，不代表模型没有接触过相关概念。",
      [
        "clip"
      ]
    ],
    [
      "3. 相似度分数取决于候选集合",
      "若候选文本只包含两个类别，模型被迫在这两个描述之间选择，不能据此证明图片里一定属于其中之一。提示措辞、类别粒度和多对象场景会影响使用方式。本文建议同时保留候选文本、图像预处理和拒识规则，并分别评估识别、计数、关系与文字读取；不要用一个分类分数代替所有视觉任务。",
      []
    ],
    [
      "4. Perceiver 用潜变量瓶颈处理输入",
      "Perceiver 使用非对称注意力将高维输入汇聚到较小潜变量集合，再在潜变量中处理。它讨论的是输入规模与内部计算之间的组织关系，与 CLIP 的图文对比目标不是同一个维度。采用瓶颈会带来信息选择问题：重要细节是否保留，需要通过具体任务与分辨率条件检验。",
      [
        "perceiver"
      ]
    ],
    [
      "5. 分辨率与分块的解析关系",
      "假设正方形图像边长为 H，块边长为 P 且整除，图像块数为 (H/P)²。取 P=16，224 像素边长产生 196 块，448 产生 784 块；若仅看图像块之间的完整注意力配对，数量分别为 38,416 与 614,656，增长 16 倍。图 2 不含分类词元、批量或通道，也不是实际 FLOPs 或显存结果。",
      [
        "vit"
      ]
    ],
    [
      "6. 选择输入预算应遵循任务",
      "读取小字与判断大轮廓，对信息保留的需求不同。本文建议先确定任务需要的最小视觉线索，再比较缩放、裁剪与分块策略。如果把高分辨率图像缩小到无法识别文字，再责备推理模块算错表格，就把输入丢失与推理错误混在了一起。评估中应保留原图与模型实际输入的对应关系。",
      []
    ],
    [
      "7. 跨模态评估要追踪整条链路",
      "一个系统可能使用图像编码器、投影器、语言模型和外部 OCR。应明确每个模块的版本与训练边界，记录哪些信息来自原图、哪些来自工具。若回答含有图中不存在的属性，应检查是感知、文本先验还是生成阶段引入；只有保留模块输入输出，错误才可能被定位。",
      []
    ],
    [
      "对齐表示并不意味着对象逐一对齐",
      "整张图片与一句文字的匹配目标，可以鼓励概念层面的对应，却不必提供像素级定位、对象绑定或完整空间关系。回答“红色物体在蓝色物体左边”需要的证据，与判断图片和“街道”一词相关并不相同。因此，在引用图文对齐研究时，应明确被证明的是哪种迁移任务，避免把相似度能力扩大为所有视觉推理能力。",
      []
    ],
    [
      "后续实验可以如何设计",
      "固定图像来源与候选任务，保留原图和实际输入。分别改变分辨率、分块或候选文本，其余条件保持一致；用细粒度任务观察丢失的是小字、局部对象还是整体关系。对经过 OCR 的管线单独记录文字转写错误，避免把 OCR 漏字全部记到后续语言推理上。所有结果应绑定所用预处理版本。",
      []
    ],
    [
      "局限与结论",
      "本文没有训练视觉模型，分块图是解析示例，未重绘任何模型胜率。ViT、CLIP 与 Perceiver 分别提供表示、监督和输入压缩思路，可以组合，也可能受任务限制。理解多模态系统，应先问输入如何变成表示，再问目标鼓励什么，最后用实际任务检验被保留的信息。",
      []
    ]
  ],
  "diagram": [
    "原始图像",
    "记录尺寸、裁剪和输入质量",
    "ViT：分块表示",
    "将图像块投影为序列",
    "CLIP：图文对比",
    "训练目标连接图像与文本表示",
    "Perceiver：输入压缩",
    "交叉注意力汇聚到潜变量",
    "下游任务接口",
    "候选文本、生成或其他读出",
    "分任务验收",
    "识别、计数、关系和文字分开"
  ],
  "figureNote": "图 1 取 P=16，H=224/448，计算图像块与块间配对数量；不含额外词元，不代表实测 FLOPs、显存或准确率。",
  "table": {
    "headers": [
      "研究线索",
      "主要设计对象",
      "不能直接替代"
    ],
    "rows": [
      [
        "ViT",
        "图像块的序列表示",
        "数据与计算条件"
      ],
      [
        "CLIP",
        "图文对比监督目标",
        "精确定位和关系测量"
      ],
      [
        "Perceiver",
        "输入到潜变量的汇聚",
        "任务信息是否完整保留"
      ]
    ]
  },
  "version": 4,
  "kind": "ai_assisted_research_guide",
  "author": "Codex（AI）",
  "published_date": "2026-10-16",
  "disclosure": "站点用户公开委托，Codex（AI）依据一手资料整理；封面由 ImageGen 生成，数据图由代码绘制；未经独立同行评审，未开展新的模型实验。",
  "limitations": "文献导读与方法分析；不是原创实验论文、正式期刊录用或专业资质认证。解析示例不代表本站或当前产品的实测表现。",
  "sources": [
    {
      "id": "vit",
      "title": "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale",
      "authors": "Dosovitskiy et al.",
      "date": "2020-10-22",
      "url": "https://arxiv.org/html/2010.11929v2",
      "locator": "v2; architecture and scaling experiments",
      "note": "Images split into patches become tokens for a Transformer; large-scale pretraining matters to results.",
      "accessed_at": "2026-10-15",
      "rights": "Original source remains under its own license; only attributed factual synthesis, no source figures or long excerpts."
    },
    {
      "id": "clip",
      "title": "Learning Transferable Visual Models From Natural Language Supervision",
      "authors": "Radford et al.",
      "date": "2021-02-26",
      "url": "https://arxiv.org/html/2103.00020v1",
      "locator": "v1; contrastive training and zero-shot transfer",
      "note": "CLIP contrasts image-text pairs; its reported training corpus contains 400 million pairs. Candidate text prompts define a zero-shot classification task.",
      "accessed_at": "2026-10-15",
      "rights": "Original source remains under its own license; only attributed factual synthesis, no source figures or long excerpts."
    },
    {
      "id": "perceiver",
      "title": "Perceiver: General Perception with Iterative Attention",
      "authors": "Jaegle et al.",
      "date": "2021-03-04",
      "url": "https://arxiv.org/abs/2103.03206v2",
      "locator": "v2 abstract; latent bottleneck architecture",
      "note": "An asymmetric cross-attention bottleneck processes high-dimensional inputs and decouples input size from latent self-attention size.",
      "accessed_at": "2026-10-15",
      "rights": "Original source remains under its own license; only attributed factual synthesis, no source figures or long excerpts."
    }
  ],
  "figures": [
    {
      "slug": "multimodal-representations",
      "role": "conceptual-cover",
      "generator": "OpenAI built-in image_gen",
      "ai_generated": true,
      "alt": "波形、玻璃平面和金属杆在共同空间相遇的多模态概念插画。",
      "original_sha256": "485b8f65b91eca9024c6a4ff9388fdb4e8855f6d05169e1e0540ee608f988658",
      "prompt": "Create exceptional premium science journal editorial COVER ART in landscape 3:2, no text, labels, numbers, logos, border or watermark. A sculptural still-life with extremely believable tactile materials and sophisticated art direction, beautiful soft cinematic lighting, precise craftsmanship, warm gold details, rich yet restrained colors. Coherent with museum sculpture photography in a contemporary research magazine, memorable at thumbnail size. Clearly conceptual art and not a new scientific result. Avoid plastic CGI, glowing brain and microchip clichés, rainbow gradients, flat vector infographics and neon cyberpunk. Topic: multimodal representation: image, sound and language entering a shared space. An elegant piece of gallery art: three distinct forms converge at a floating transparent oblique prism — a translucent rectangular blue-green glass pane with subtle mosaic-like facets, an ivory ceramic continuous wave ribbon, and a few slender brass vertical rods. They meet without becoming cluttered. Backdrop deep dusty teal, matte stone foreground, coral accent caustic, warm soft gold light from right, varied materials clearly distinguish modalities, dramatic yet calm composition with negative space.",
      "files": [
        {
          "file": "assets/covers/multimodal-representations.png",
          "width": 1280,
          "height": 853,
          "bytes": 1433502,
          "sha256": "052af8ac4b7ee0160a48c0c85bb26e3fcf82104c844d204f24bf07b6c1d1d857",
          "url": "/guides/assets/v2-multimodal-representations-052af8ac4b7e.png"
        },
        {
          "file": "assets/covers/multimodal-representations-480.png",
          "width": 480,
          "height": 320,
          "bytes": 211698,
          "sha256": "42da4d4452d5ce3c8b4b2c9854cd2c27329dcc266d0e8d7a769a35da1b188608",
          "url": "/guides/assets/v2-multimodal-representations-480-42da4d4452d5.png"
        }
      ],
      "url": "/guides/assets/v2-multimodal-representations-052af8ac4b7e.png",
      "thumbnail_url": "/guides/assets/v2-multimodal-representations-480-42da4d4452d5.png"
    },
    {
      "slug": "multimodal-representations",
      "role": "data-chart",
      "file": "assets/charts/multimodal-representations.png",
      "width": 1800,
      "height": 1125,
      "sha256": "95c6ef66b6b525156b6ae847aa4b002e08face8a111d9a93b14d401c2bdf3cac",
      "bytes": 126328,
      "alt": "图表：取 P=16，H=224/448，计算图像块与块间配对数量；不含额外词元，不代表实测 FLOPs、显存或准确率。",
      "data": {
        "kind": "analytical",
        "patch_side": 16,
        "image_side": [
          224,
          448
        ],
        "patches": [
          196,
          784
        ],
        "attention_pairs": [
          38416,
          614656
        ],
        "excludes": "CLS and other tokens; not FLOPs or observed memory"
      },
      "data_source_sha256": "963323d4a7a0cad4aebd986f09a55251a7c21f68c757d21b76245a128675e13e",
      "ai_generated": true,
      "disclosure": "AI-assisted deterministic rendering of unchanged v1 data; not a new experiment.",
      "url": "/guides/assets/v2-multimodal-representations-95c6ef66b6b5.png"
    }
  ],
  "peer_reviewed": false,
  "previous_version": "/guides/multimodal-representations/v3/",
  "revision": "Mathematical typography only; scientific prose, numeric data, sources and figures are unchanged.",
  "source_language": "zh-Hans",
  "original_version": 2,
  "original_article_sha256": "9fef8c68b796d122f800d593baf52b202dfd046d2e204e9a70b58e2f1d2bb3c3",
  "interface_localizations": {
    "en": {
      "title": "From image patches to image–text alignment: multimodal representations",
      "dek": "Compare visual backbones, contrastive objectives and input compression in ViT and CLIP.",
      "translation_method": "AI-assisted translation",
      "review_status": "not independently language-reviewed"
    },
    "zh": {
      "title": "从图像分块到图文对齐：ViT、CLIP 与多模态表示的边界",
      "dek": "区分视觉骨干、跨模态训练目标和输入压缩，说明“图文相似”为什么不能直接等同于完整场景理解。",
      "translation_method": "original",
      "review_status": "original"
    },
    "es": {
      "title": "De los parches a la alineación entre imagen y texto: representaciones multimodales",
      "dek": "Comparar redes visuales, objetivos contrastivos y compresión de entradas en ViT y CLIP.",
      "translation_method": "AI-assisted translation",
      "review_status": "not independently language-reviewed"
    },
    "ja": {
      "title": "画像パッチから画像と言語の整合へ：マルチモーダル表現の境界",
      "dek": "ViT と CLIP の視覚基盤、対照学習の目的、入力圧縮を比較する。",
      "translation_method": "AI-assisted translation",
      "review_status": "not independently language-reviewed"
    },
    "ar": {
      "title": "من رقع الصور إلى مواءمة الصورة والنص: التمثيلات متعددة الوسائط",
      "dek": "مقارنة البنى البصرية وأهداف التعلم التقابلي وضغط المدخلات في ViT وCLIP.",
      "translation_method": "AI-assisted translation",
      "review_status": "not independently language-reviewed"
    }
  },
  "presentation_revision": {
    "kind": "scientific_math_typography",
    "source_head": "57294bb73b50b2da740f2e8ac63a3d4d7f9a6843",
    "source_v3_sha256": "c09cde3f78e6bcdd8613243419a431c97a340453dfd2f4dd4ed2ae71840b18bc",
    "annotations_sha256": "2d57369e64d81dc061ddf3bb16d964ebe272b0a04a6a351b888d3470592554f6"
  }
}