{
  "id": 187366,
  "title": "What is the Raster Class?",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/187366",
  "author_name": "",
  "post_date": "2020-09-28T16:42:27.689074500Z",
  "votes": 1,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I went through a couple of kernels and found the unit 'RASTERIZATION' everywhere!! Can anyone please explain what is this RASTER class and why is it so essential in the L5Kit module? Moreover, are there any other equally important classes present? Understanding the L5Kit is proving to be a bit difficult! Thank you</p>",
  "messages": [
    {
      "id": "1030407",
      "postDate": "09/28/2020 16:42:27",
      "content": "<p>I went through a couple of kernels and found the unit 'RASTERIZATION' everywhere!! Can anyone please explain what is this RASTER class and why is it so essential in the L5Kit module? Moreover, are there any other equally important classes present? Understanding the L5Kit is proving to be a bit difficult! Thank you</p>",
      "rawMarkdown": "I went through a couple of kernels and found the unit 'RASTERIZATION' everywhere!! Can anyone please explain what is this RASTER class and why is it so essential in the L5Kit module? Moreover, are there any other equally important classes present? Understanding the L5Kit is proving to be a bit difficult! Thank you",
      "votes": null
    },
    {
      "id": "1030547",
      "postDate": "09/28/2020 18:03:45",
      "content": "<p>Rasterization is a process of creating images from other objects. For example below is a typical image that we get, with 25 channels, channel by channel view. First 11 images are rasterizations of other agents history, next 11 images are the agent under consideration itself, and the last 3 is a sematic map rasterization. </p>\n<p><img src=\"https://imgur.com/7lQ9CmU.png\" alt=\"rasterization\"></p>",
      "rawMarkdown": "Rasterization is a process of creating images from other objects. For example below is a typical image that we get, with 25 channels, channel by channel view. First 11 images are rasterizations of other agents history, next 11 images are the agent under consideration itself, and the last 3 is a sematic map rasterization. \n\n![rasterization](https://imgur.com/7lQ9CmU.png)",
      "votes": null
    },
    {
      "id": "1030651",
      "postDate": "09/28/2020 20:02:07",
      "content": "<p>The official paper to the dataset explains a lot of things, have a look:<br>\n<a href=\"https://arxiv.org/pdf/2006.14480.pdf\" target=\"_blank\">https://arxiv.org/pdf/2006.14480.pdf</a></p>",
      "rawMarkdown": "The official paper to the dataset explains a lot of things, have a look:\nhttps://arxiv.org/pdf/2006.14480.pdf",
      "votes": null
    },
    {
      "id": "1033118",
      "postDate": "09/30/2020 17:15:45",
      "content": "<p>thank you for the information!! really helpful</p>",
      "rawMarkdown": "thank you for the information!! really helpful",
      "votes": null
    },
    {
      "id": "1033119",
      "postDate": "09/30/2020 17:16:01",
      "content": "<p>thank you so much!</p>\n<p>All the best~~</p>",
      "rawMarkdown": "thank you so much!\n\nAll the best~~",
      "votes": null
    },
    {
      "id": "1069670",
      "postDate": "11/04/2020 18:30:44",
      "content": "<p><a href=\"https://www.kaggle.com/zaharch\" target=\"_blank\">@zaharch</a> </p>\n<blockquote>\n  <p>First 11 images are rasterizations of other agents history,</p>\n</blockquote>\n<p>While I can understand the use of present location of other agents, what information do the histories (the other 10 channels) of other agents provide to the network ? Is it that it provides additional source of data for \"learning historical speed of agents?\" </p>",
      "rawMarkdown": "zaharch \n>First 11 images are rasterizations of other agents history,\n\nWhile I can understand the use of present location of other agents, what information do the histories (the other 10 channels) of other agents provide to the network ? Is it that it provides additional source of data for \"learning historical speed of agents?\"",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1030547,
      "author_name": "zaharch",
      "author_url": "",
      "post_date": "09/28/2020 18:03:45",
      "content": "<p>Rasterization is a process of creating images from other objects. For example below is a typical image that we get, with 25 channels, channel by channel view. First 11 images are rasterizations of other agents history, next 11 images are the agent under consideration itself, and the last 3 is a sematic map rasterization. </p>\n<p><img src=\"https://imgur.com/7lQ9CmU.png\" alt=\"rasterization\"></p>",
      "votes": null,
      "replies": [
        {
          "id": 1033118,
          "author_name": "fireheart7",
          "author_url": "",
          "post_date": "09/30/2020 17:15:45",
          "content": "<p>thank you for the information!! really helpful</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1069670,
          "author_name": "watzisname",
          "author_url": "",
          "post_date": "11/04/2020 18:30:44",
          "content": "<p><a href=\"https://www.kaggle.com/zaharch\" target=\"_blank\">@zaharch</a> </p>\n<blockquote>\n  <p>First 11 images are rasterizations of other agents history,</p>\n</blockquote>\n<p>While I can understand the use of present location of other agents, what information do the histories (the other 10 channels) of other agents provide to the network ? Is it that it provides additional source of data for \"learning historical speed of agents?\" </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1030651,
      "author_name": "aliabdin1",
      "author_url": "",
      "post_date": "09/28/2020 20:02:07",
      "content": "<p>The official paper to the dataset explains a lot of things, have a look:<br>\n<a href=\"https://arxiv.org/pdf/2006.14480.pdf\" target=\"_blank\">https://arxiv.org/pdf/2006.14480.pdf</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1033119,
          "author_name": "fireheart7",
          "author_url": "",
          "post_date": "09/30/2020 17:16:01",
          "content": "<p>thank you so much!</p>\n<p>All the best~~</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1030407": "I went through a couple of kernels and found the unit 'RASTERIZATION' everywhere!! Can anyone please explain what is this RASTER class and why is it so essential in the L5Kit module? Moreover, are there any other equally important classes present? Understanding the L5Kit is proving to be a bit difficult! Thank you",
    "1030547": "Rasterization is a process of creating images from other objects. For example below is a typical image that we get, with 25 channels, channel by channel view. First 11 images are rasterizations of other agents history, next 11 images are the agent under consideration itself, and the last 3 is a sematic map rasterization. \n\n![rasterization](https://imgur.com/7lQ9CmU.png)",
    "1030651": "The official paper to the dataset explains a lot of things, have a look:\nhttps://arxiv.org/pdf/2006.14480.pdf",
    "1033118": "thank you for the information!! really helpful",
    "1033119": "thank you so much!\n\nAll the best~~",
    "1069670": "zaharch \n>First 11 images are rasterizations of other agents history,\n\nWhile I can understand the use of present location of other agents, what information do the histories (the other 10 channels) of other agents provide to the network ? Is it that it provides additional source of data for \"learning historical speed of agents?\""
  },
  "source": "meta"
}