{
  "id": 49245,
  "title": "Same device classification for a pair of images?",
  "url": "/competitions/sp-society-camera-model-identification/discussion/49245",
  "author_name": "",
  "post_date": "2018-02-08T12:10:13.054964100Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>For each camera type, only a single device is used both for training and testing images. Even though training device is different from the testing device, I imagine training a siamese-like network to say whether images were captured with the same device could help in grouping and averaging the predictions for test set images.</p>\n\n<p>Has anyone tried something like this?</p>",
  "messages": [
    {
      "id": "279634",
      "postDate": "02/08/2018 12:10:13",
      "content": "<p>For each camera type, only a single device is used both for training and testing images. Even though training device is different from the testing device, I imagine training a siamese-like network to say whether images were captured with the same device could help in grouping and averaging the predictions for test set images.</p>\n\n<p>Has anyone tried something like this?</p>",
      "rawMarkdown": "For each camera type, only a single device is used both for training and testing images. Even though training device is different from the testing device, I imagine training a siamese-like network to say whether images were captured with the same device could help in grouping and averaging the predictions for test set images.\n\nHas anyone tried something like this?",
      "votes": null
    },
    {
      "id": "279668",
      "postDate": "02/08/2018 14:04:46",
      "content": "<p>I tried this <a href=\"https://github.com/andrewlewis/camera-id\">https://github.com/andrewlewis/camera-id</a> a bit (no training involved), but results didn't seem too good.</p>",
      "rawMarkdown": "I tried this https://github.com/andrewlewis/camera-id a bit (no training involved), but results didn't seem too good.",
      "votes": null
    },
    {
      "id": "279702",
      "postDate": "02/08/2018 15:06:15",
      "content": "<p>We tried this, but results are very bad (maybe there are some bugs in our code). We will check the source code again and upload to github.</p>",
      "rawMarkdown": "We tried this, but results are very bad (maybe there are some bugs in our code). We will check the source code again and upload to github.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 279668,
      "author_name": "bobutis",
      "author_url": "",
      "post_date": "02/08/2018 14:04:46",
      "content": "<p>I tried this <a href=\"https://github.com/andrewlewis/camera-id\">https://github.com/andrewlewis/camera-id</a> a bit (no training involved), but results didn't seem too good.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 279702,
      "author_name": "mathormad",
      "author_url": "",
      "post_date": "02/08/2018 15:06:15",
      "content": "<p>We tried this, but results are very bad (maybe there are some bugs in our code). We will check the source code again and upload to github.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "279634": "For each camera type, only a single device is used both for training and testing images. Even though training device is different from the testing device, I imagine training a siamese-like network to say whether images were captured with the same device could help in grouping and averaging the predictions for test set images.\n\nHas anyone tried something like this?",
    "279668": "I tried this https://github.com/andrewlewis/camera-id a bit (no training involved), but results didn't seem too good.",
    "279702": "We tried this, but results are very bad (maybe there are some bugs in our code). We will check the source code again and upload to github."
  },
  "source": "meta"
}