{
  "docId": "019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "docSlug": "bf350dd574c19997",
  "documentTitle": "2021 Air Street Capital The State of AI Report 2021",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 33,
  "pageCount": 188,
  "prevPage": 32,
  "nextPage": 34,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "overcrowded",
  "nDataPoints": 20,
  "notes": "Includes a comparison diagram of existing methods vs. ClipBERT and a performance table.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "flowchart",
    "data_table",
    "logo_grid"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-fd5012384ae3/33",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3#slide-33",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "To solve video-and-language (V&L) tasks like video captioning, ClipBERT only uses a few sparsely sampled short clips. It still outperforms existing methods that exploit full-length videos.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d26-b00b3c5766d7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.35,
        "x": 0.62,
        "y": 0.32
      },
      "kind": "image",
      "text": "Diagram comparing existing methods (dense sampling) vs. ClipBERT (sparse sampling)",
      "attrs": null,
      "subkind": "infographic",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "830976e7-b978-424c-9efc-214416df7fa3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.55,
        "x": 0.05,
        "y": 0.32
      },
      "kind": "list",
      "text": "The usual approach to solve video-and-language tasks is to use separate task-agnostic encoders for videos and images, then use the resulting features to teach a neural network the task at hand.\nA natural improvement of this process would be end-to-end learning of vision and text encoders. But due to the length of the video clips, this is usually computationally unaffordable.\nSurprisingly, researchers show that with end-to-end learning, one only needs a few samples of a video to outperform existing methods which use full-length videos. They also verify that ClipBERT performs better with sparse random sampling than with dense uniform sampling.\nClipBERT surpasses SOTA methods on datasets for text-to-video retrieval and video QA, including MSRVTT, DiDeMo and TGIF-QA.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "df486c38-026b-4fa6-8df5-093e2fa97704",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "MSRVTT-QA Acc.: 36.67",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d26-b6b89dc275c7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.55,
        "x": 0.05,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "To solve video-and-language (V&L) tasks like video captioning, ClipBERT only uses a few sparsely sampled short clips. It still outperforms existing methods that exploit full-length videos.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "79b25581-c607-4642-ae1d-92a590ee68d9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.35,
        "x": 0.62,
        "y": 0.7
      },
      "kind": "table",
      "text": "Table 4: Sparse random sampling vs. dense uniform sampling. All models use Ntest=16 clips for inference.",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "4cdccbc5-f912-48e1-bd28-15ec895e1837",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "Less is more: watching a few clips is enough to learn how to caption a video",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d5aefcda-7938-4336-95af-00b56ff9fbe9",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Big Idea Formula",
      "slug": "big-idea-formula",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "8dd00380-6621-49ea-8149-859feb07a501",
      "evidence": "The slide presents a novel approach to video-and-language tasks, highlighting the idea that less data can be more effective.",
      "confidence": 0.7
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 152,
      "from": 5,
      "beatId": "019dd95a-0682-776c-8e35-0affbadec38e",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Definitions p.5-6, Exec Summary p.7, then Sections 1-3 catalog evidence.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 152,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e35-1b3ee3a1c012",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Three sections accumulating evidence of accelerating progress.",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}