{
  "id": 37455,
  "title": "Photo without a car",
  "url": "/competitions/carvana-image-masking-challenge/discussion/37455",
  "author_name": "",
  "post_date": "2017-08-02T19:02:21.153690100Z",
  "votes": 3,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi, is it possible to get several photos of the place without any car?</p>",
  "messages": [
    {
      "id": "209593",
      "postDate": "08/02/2017 19:02:21",
      "content": "<p>Hi, is it possible to get several photos of the place without any car?</p>",
      "rawMarkdown": "Hi, is it possible to get several photos of the place without any car?",
      "votes": null
    },
    {
      "id": "209601",
      "postDate": "08/02/2017 19:25:23",
      "content": "<p>I believe they stated that images without cars aren't taken.  One way I've tried to get around this is to invert the mask, and then take the difference of the 16 images for a given car.  This has the net effect of rotating the background 360 degrees and what you're left with is the background and the minimum car mask that has an overlap only where the car exists in all 16 images. \nYou could extend the concept to all cars. Be careful though, I believe there are three different backgrounds.\nHere's an example of what I mean for just one car rotated.  <img src=\"http://imgur.com/a/nIlln\" alt=\"Inverted mask rotated\" title=\"\"></p>",
      "rawMarkdown": "I believe they stated that images without cars aren't taken.  One way I've tried to get around this is to invert the mask, and then take the difference of the 16 images for a given car.  This has the net effect of rotating the background 360 degrees and what you're left with is the background and the minimum car mask that has an overlap only where the car exists in all 16 images. \nYou could extend the concept to all cars. Be careful though, I believe there are three different backgrounds.\nHere's an example of what I mean for just one car rotated.  ![Inverted mask rotated][1]\n\n\n  [1]: http://imgur.com/a/nIlln",
      "votes": null
    },
    {
      "id": "209607",
      "postDate": "08/02/2017 19:39:43",
      "content": "<p>I was able to dig up one photo from a couple of years ago. I will need to discuss with the Kaggle team how to release it fairly.</p>\n\n<p>It's important to note that there are multiple physical locations, so one reference frame will not align perfectly all photos. Occasionally, when a studio is serviced, things like the paint on the floor, camera's position, and lighting color temperature can change slightly.</p>\n\n<p>If we do release a reference frame, it may not be of much value.</p>",
      "rawMarkdown": "I was able to dig up one photo from a couple of years ago. I will need to discuss with the Kaggle team how to release it fairly.\n\nIt's important to note that there are multiple physical locations, so one reference frame will not align perfectly all photos. Occasionally, when a studio is serviced, things like the paint on the floor, camera's position, and lighting color temperature can change slightly.\n\nIf we do release a reference frame, it may not be of much value.",
      "votes": null
    },
    {
      "id": "209632",
      "postDate": "08/02/2017 21:39:26",
      "content": "<p>Also the position of the camera changes and possibly the focal length when fitting different sized cars.</p>\n\n<p>A “clean plate” could be made in a 3d package like maya by baking camera projections on the ground with the masks and composited together. </p>\n\n<p>You would have to keep in mind that the floor is rotating and rotate the baked projections before you comp them together </p>\n\n<p>The wall doesn’t have much info besides the text and could be manually created.</p>",
      "rawMarkdown": "Also the position of the camera changes and possibly the focal length when fitting different sized cars.\n\nA “clean plate” could be made in a 3d package like maya by baking camera projections on the ground with the masks and composited together. \n\nYou would have to keep in mind that the floor is rotating and rotate the baked projections before you comp them together \n\nThe wall doesn’t have much info besides the text and could be manually created.",
      "votes": null
    },
    {
      "id": "209691",
      "postDate": "08/03/2017 03:33:07",
      "content": "<p>What is a good way of merging/blending all of these image projections into a single image?</p>\n\n<p><img src=\"http://i.imgur.com/vf6lDB4.gifv\" alt=\"My try at projecting the ground\" title=\"\"></p>",
      "rawMarkdown": "What is a good way of merging/blending all of these image projections into a single image?\n\n![My try at projecting the ground][1]\n  [1]: http://i.imgur.com/vf6lDB4.gifv",
      "votes": null
    },
    {
      "id": "209739",
      "postDate": "08/03/2017 08:43:34",
      "content": "<p>The focal length may vary slightly, but by studio rather than by vehicle. In a single studio, the focal length is static. After the first one was built, I think the size of the studio, camera position, and focal length were adjusted slightly to handle larger vehicles.</p>",
      "rawMarkdown": "The focal length may vary slightly, but by studio rather than by vehicle. In a single studio, the focal length is static. After the first one was built, I think the size of the studio, camera position, and focal length were adjusted slightly to handle larger vehicles.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 209601,
      "author_name": "tivfrvqhs5",
      "author_url": "",
      "post_date": "08/02/2017 19:25:23",
      "content": "<p>I believe they stated that images without cars aren't taken.  One way I've tried to get around this is to invert the mask, and then take the difference of the 16 images for a given car.  This has the net effect of rotating the background 360 degrees and what you're left with is the background and the minimum car mask that has an overlap only where the car exists in all 16 images. \nYou could extend the concept to all cars. Be careful though, I believe there are three different backgrounds.\nHere's an example of what I mean for just one car rotated.  <img src=\"http://imgur.com/a/nIlln\" alt=\"Inverted mask rotated\" title=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 209607,
      "author_name": "brianshaler",
      "author_url": "",
      "post_date": "08/02/2017 19:39:43",
      "content": "<p>I was able to dig up one photo from a couple of years ago. I will need to discuss with the Kaggle team how to release it fairly.</p>\n\n<p>It's important to note that there are multiple physical locations, so one reference frame will not align perfectly all photos. Occasionally, when a studio is serviced, things like the paint on the floor, camera's position, and lighting color temperature can change slightly.</p>\n\n<p>If we do release a reference frame, it may not be of much value.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 209632,
      "author_name": "bhowiebkr",
      "author_url": "",
      "post_date": "08/02/2017 21:39:26",
      "content": "<p>Also the position of the camera changes and possibly the focal length when fitting different sized cars.</p>\n\n<p>A “clean plate” could be made in a 3d package like maya by baking camera projections on the ground with the masks and composited together. </p>\n\n<p>You would have to keep in mind that the floor is rotating and rotate the baked projections before you comp them together </p>\n\n<p>The wall doesn’t have much info besides the text and could be manually created.</p>",
      "votes": null,
      "replies": [
        {
          "id": 209691,
          "author_name": "bhowiebkr",
          "author_url": "",
          "post_date": "08/03/2017 03:33:07",
          "content": "<p>What is a good way of merging/blending all of these image projections into a single image?</p>\n\n<p><img src=\"http://i.imgur.com/vf6lDB4.gifv\" alt=\"My try at projecting the ground\" title=\"\"></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 209739,
          "author_name": "brianshaler",
          "author_url": "",
          "post_date": "08/03/2017 08:43:34",
          "content": "<p>The focal length may vary slightly, but by studio rather than by vehicle. In a single studio, the focal length is static. After the first one was built, I think the size of the studio, camera position, and focal length were adjusted slightly to handle larger vehicles.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "209593": "Hi, is it possible to get several photos of the place without any car?",
    "209601": "I believe they stated that images without cars aren't taken.  One way I've tried to get around this is to invert the mask, and then take the difference of the 16 images for a given car.  This has the net effect of rotating the background 360 degrees and what you're left with is the background and the minimum car mask that has an overlap only where the car exists in all 16 images. \nYou could extend the concept to all cars. Be careful though, I believe there are three different backgrounds.\nHere's an example of what I mean for just one car rotated.  ![Inverted mask rotated][1]\n\n\n  [1]: http://imgur.com/a/nIlln",
    "209607": "I was able to dig up one photo from a couple of years ago. I will need to discuss with the Kaggle team how to release it fairly.\n\nIt's important to note that there are multiple physical locations, so one reference frame will not align perfectly all photos. Occasionally, when a studio is serviced, things like the paint on the floor, camera's position, and lighting color temperature can change slightly.\n\nIf we do release a reference frame, it may not be of much value.",
    "209632": "Also the position of the camera changes and possibly the focal length when fitting different sized cars.\n\nA “clean plate” could be made in a 3d package like maya by baking camera projections on the ground with the masks and composited together. \n\nYou would have to keep in mind that the floor is rotating and rotate the baked projections before you comp them together \n\nThe wall doesn’t have much info besides the text and could be manually created.",
    "209691": "What is a good way of merging/blending all of these image projections into a single image?\n\n![My try at projecting the ground][1]\n  [1]: http://i.imgur.com/vf6lDB4.gifv",
    "209739": "The focal length may vary slightly, but by studio rather than by vehicle. In a single studio, the focal length is static. After the first one was built, I think the size of the studio, camera position, and focal length were adjusted slightly to handle larger vehicles."
  },
  "source": "meta"
}