{
  "id": 120721,
  "title": "Are upper parts of images useful?",
  "url": "/competitions/pku-autonomous-driving/discussion/120721",
  "author_name": "",
  "post_date": "2019-12-08T09:48:31.420855800Z",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>You can see car distribution in <a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a></p>",
  "messages": [
    {
      "id": "690278",
      "postDate": "12/08/2019 09:48:31",
      "content": "<p>You can see car distribution in <a href=\"https://www.kaggle.com/hocop1/centernet-baseline\">https://www.kaggle.com/hocop1/centernet-baseline</a></p>",
      "rawMarkdown": "You can see car distribution in https://www.kaggle.com/hocop1/centernet-baseline",
      "votes": null
    },
    {
      "id": "690390",
      "postDate": "12/08/2019 13:46:30",
      "content": "<p>Hi <a href=\"/a1850961785\">@a1850961785</a> ,</p>\n\n<p>The answer is NO: pixels from line 0 to 1369 don't contain car in the training dataset.</p>\n\n<p>I determined this by getting masks of all cars from the training dataset and displayed them into a single image:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1181310%2F4a8be655ff723b4ab246b7afeaed31a2%2Fis_upper_useful.png?generation=1575812188035650&amp;alt=media\" alt=\"\"></p>\n\n<p>To reproduce you can get inspiration from my <a href=\"https://www.kaggle.com/ebouteillon/augmented-reality\">augmented-reality notebook</a> to create masks and with the car mapping from <a href=\"https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py\">here</a>.</p>",
      "rawMarkdown": "Hi @a1850961785 ,\n\nThe answer is NO: pixels from line 0 to 1369 don't contain car in the training dataset.\n\nI determined this by getting masks of all cars from the training dataset and displayed them into a single image:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1181310%2F4a8be655ff723b4ab246b7afeaed31a2%2Fis_upper_useful.png?generation=1575812188035650&amp;alt=media)\n\nTo reproduce you can get inspiration from my [augmented-reality notebook](https://www.kaggle.com/ebouteillon/augmented-reality) to create masks and with the car mapping from [here](https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py).",
      "votes": null
    },
    {
      "id": "690724",
      "postDate": "12/09/2019 02:26:02",
      "content": "<p>Thanks for you answer. \nI'm finding a way to get rid of the upper parts for better performance.</p>",
      "rawMarkdown": "Thanks for you answer. \nI'm finding a way to get rid of the upper parts for better performance.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 690390,
      "author_name": "ebouteillon",
      "author_url": "",
      "post_date": "12/08/2019 13:46:30",
      "content": "<p>Hi <a href=\"/a1850961785\">@a1850961785</a> ,</p>\n\n<p>The answer is NO: pixels from line 0 to 1369 don't contain car in the training dataset.</p>\n\n<p>I determined this by getting masks of all cars from the training dataset and displayed them into a single image:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1181310%2F4a8be655ff723b4ab246b7afeaed31a2%2Fis_upper_useful.png?generation=1575812188035650&amp;alt=media\" alt=\"\"></p>\n\n<p>To reproduce you can get inspiration from my <a href=\"https://www.kaggle.com/ebouteillon/augmented-reality\">augmented-reality notebook</a> to create masks and with the car mapping from <a href=\"https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py\">here</a>.</p>",
      "votes": null,
      "replies": [
        {
          "id": 690724,
          "author_name": "a1850961785",
          "author_url": "",
          "post_date": "12/09/2019 02:26:02",
          "content": "<p>Thanks for you answer. \nI'm finding a way to get rid of the upper parts for better performance.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "690278": "You can see car distribution in https://www.kaggle.com/hocop1/centernet-baseline",
    "690390": "Hi @a1850961785 ,\n\nThe answer is NO: pixels from line 0 to 1369 don't contain car in the training dataset.\n\nI determined this by getting masks of all cars from the training dataset and displayed them into a single image:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1181310%2F4a8be655ff723b4ab246b7afeaed31a2%2Fis_upper_useful.png?generation=1575812188035650&amp;alt=media)\n\nTo reproduce you can get inspiration from my [augmented-reality notebook](https://www.kaggle.com/ebouteillon/augmented-reality) to create masks and with the car mapping from [here](https://raw.githubusercontent.com/ApolloScapeAuto/dataset-api/master/car_instance/car_models.py).",
    "690724": "Thanks for you answer. \nI'm finding a way to get rid of the upper parts for better performance."
  },
  "source": "meta"
}