{
  "id": 198845,
  "title": "lane direction in semantic rasterizer",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/198845",
  "author_name": "",
  "post_date": "2020-11-23T09:28:55.456753700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>has anyone used the lane direction in the semantic rasterizer with success?</p>\n<p>I retrieved the lane direction in the semantic rasterizer (render_semantic_map method) with:<br>\n<code>lane.orientation_in_parent_segment</code></p>\n<p>Then i colored the forward lanes with more green, the backward lanes with more red and both ways with more red and green. The first image i tested looked exactly like i hoped it would look. But checking more images i think the forward/backward direction is with respect to the av and not the current agent. Also on Motorways all lanes seem backwards. So in the end i just colored the forward lanes with some more green, but it didn't improve my model.</p>\n<p>I am just curious if someone got reasonable data from the lane direction. I totally understand if you answer this after the competition end, but i am curious.</p>",
  "messages": [
    {
      "id": "1088060",
      "postDate": "11/23/2020 09:28:55",
      "content": "<p>Hi,</p>\n<p>has anyone used the lane direction in the semantic rasterizer with success?</p>\n<p>I retrieved the lane direction in the semantic rasterizer (render_semantic_map method) with:<br>\n<code>lane.orientation_in_parent_segment</code></p>\n<p>Then i colored the forward lanes with more green, the backward lanes with more red and both ways with more red and green. The first image i tested looked exactly like i hoped it would look. But checking more images i think the forward/backward direction is with respect to the av and not the current agent. Also on Motorways all lanes seem backwards. So in the end i just colored the forward lanes with some more green, but it didn't improve my model.</p>\n<p>I am just curious if someone got reasonable data from the lane direction. I totally understand if you answer this after the competition end, but i am curious.</p>",
      "rawMarkdown": "Hi,\n\nhas anyone used the lane direction in the semantic rasterizer with success?\n\nI retrieved the lane direction in the semantic rasterizer (render_semantic_map method) with:\n`lane.orientation_in_parent_segment`\n\nThen i colored the forward lanes with more green, the backward lanes with more red and both ways with more red and green. The first image i tested looked exactly like i hoped it would look. But checking more images i think the forward/backward direction is with respect to the av and not the current agent. Also on Motorways all lanes seem backwards. So in the end i just colored the forward lanes with some more green, but it didn't improve my model.\n\nI am just curious if someone got reasonable data from the lane direction. I totally understand if you answer this after the competition end, but i am curious.",
      "votes": null
    },
    {
      "id": "1088395",
      "postDate": "11/23/2020 15:14:45",
      "content": "<p>I did not try to do that, but if you have lanes direction, you should be able to calculate an angle of the direction vector and use that angle as a hue in HSV color model.  This way opposite directions will always have completely opposite colors. </p>",
      "rawMarkdown": "I did not try to do that, but if you have lanes direction, you should be able to calculate an angle of the direction vector and use that angle as a hue in HSV color model.  This way opposite directions will always have completely opposite colors.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1088395,
      "author_name": "szacho",
      "author_url": "",
      "post_date": "11/23/2020 15:14:45",
      "content": "<p>I did not try to do that, but if you have lanes direction, you should be able to calculate an angle of the direction vector and use that angle as a hue in HSV color model.  This way opposite directions will always have completely opposite colors. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1088060": "Hi,\n\nhas anyone used the lane direction in the semantic rasterizer with success?\n\nI retrieved the lane direction in the semantic rasterizer (render_semantic_map method) with:\n`lane.orientation_in_parent_segment`\n\nThen i colored the forward lanes with more green, the backward lanes with more red and both ways with more red and green. The first image i tested looked exactly like i hoped it would look. But checking more images i think the forward/backward direction is with respect to the av and not the current agent. Also on Motorways all lanes seem backwards. So in the end i just colored the forward lanes with some more green, but it didn't improve my model.\n\nI am just curious if someone got reasonable data from the lane direction. I totally understand if you answer this after the competition end, but i am curious.",
    "1088395": "I did not try to do that, but if you have lanes direction, you should be able to calculate an angle of the direction vector and use that angle as a hue in HSV color model.  This way opposite directions will always have completely opposite colors."
  },
  "source": "meta"
}