{
  "id": 319066,
  "title": "Hey there. Question about cropped public dataset and the final test dataset",
  "url": "/competitions/happy-whale-and-dolphin/discussion/319066",
  "author_name": "",
  "post_date": "2022-04-15T08:28:31.509459400Z",
  "votes": 3,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I've been training and testing my model with cropped dataset. I realize that the final test dataset won't be cropped. Should I add in my codes a function to crop the final test dataset before inferencing?</p>",
  "messages": [
    {
      "id": "1756088",
      "postDate": "04/15/2022 08:28:31",
      "content": "<p>I've been training and testing my model with cropped dataset. I realize that the final test dataset won't be cropped. Should I add in my codes a function to crop the final test dataset before inferencing?</p>",
      "rawMarkdown": "I've been training and testing my model with cropped dataset. I realize that the final test dataset won't be cropped. Should I add in my codes a function to crop the final test dataset before inferencing?",
      "votes": null
    },
    {
      "id": "1756109",
      "postDate": "04/15/2022 08:52:44",
      "content": "<p>Usually yes, you should predict the same preprocessing (without some hard augmentation, or use it in TTA) like in training</p>",
      "rawMarkdown": "Usually yes, you should predict the same preprocessing (without some hard augmentation, or use it in TTA) like in training",
      "votes": null
    },
    {
      "id": "1756115",
      "postDate": "04/15/2022 09:01:06",
      "content": "<p>Mind if I ask a question further into it? How am I supposed to figure out the final test dataset folder location  when I have no clue?</p>",
      "rawMarkdown": "Mind if I ask a question further into it? How am I supposed to figure out the final test dataset folder location  when I have no clue?",
      "votes": null
    },
    {
      "id": "1756232",
      "postDate": "04/15/2022 10:22:03",
      "content": "<p>What do you mean under \"final test dataset\", you already have final test dataset, 28k images</p>",
      "rawMarkdown": "What do you mean under \"final test dataset\", you already have final test dataset, 28k images",
      "votes": null
    },
    {
      "id": "1756237",
      "postDate": "04/15/2022 10:29:06",
      "content": "<p>I thought we're going to test another training set that consists of 88k images (28k:24% = x:76%) . Oh wait…. 76% means the other 76% of THE test dataset? So the current leaderboard scores are evaluated only on 24% of the test dataset result that we submit?</p>",
      "rawMarkdown": "I thought we're going to test another training set that consists of 88k images (28k:24% = x:76%) . Oh wait.... 76% means the other 76% of THE test dataset? So the current leaderboard scores are evaluated only on 24% of the test dataset result that we submit?",
      "votes": null
    },
    {
      "id": "1756238",
      "postDate": "04/15/2022 10:29:58",
      "content": "<p>Oh my god. I was totally mistaken. Thank you so much!</p>",
      "rawMarkdown": "Oh my god. I was totally mistaken. Thank you so much!",
      "votes": null
    },
    {
      "id": "1756240",
      "postDate": "04/15/2022 10:30:13",
      "content": "<p>exactly, 24% of 28k images is public part</p>",
      "rawMarkdown": "exactly, 24% of 28k images is public part",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1756109,
      "author_name": "kwentar",
      "author_url": "",
      "post_date": "04/15/2022 08:52:44",
      "content": "<p>Usually yes, you should predict the same preprocessing (without some hard augmentation, or use it in TTA) like in training</p>",
      "votes": null,
      "replies": [
        {
          "id": 1756115,
          "author_name": "cheulkay",
          "author_url": "",
          "post_date": "04/15/2022 09:01:06",
          "content": "<p>Mind if I ask a question further into it? How am I supposed to figure out the final test dataset folder location  when I have no clue?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1756232,
          "author_name": "kwentar",
          "author_url": "",
          "post_date": "04/15/2022 10:22:03",
          "content": "<p>What do you mean under \"final test dataset\", you already have final test dataset, 28k images</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1756237,
          "author_name": "cheulkay",
          "author_url": "",
          "post_date": "04/15/2022 10:29:06",
          "content": "<p>I thought we're going to test another training set that consists of 88k images (28k:24% = x:76%) . Oh wait…. 76% means the other 76% of THE test dataset? So the current leaderboard scores are evaluated only on 24% of the test dataset result that we submit?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1756238,
          "author_name": "cheulkay",
          "author_url": "",
          "post_date": "04/15/2022 10:29:58",
          "content": "<p>Oh my god. I was totally mistaken. Thank you so much!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1756240,
          "author_name": "kwentar",
          "author_url": "",
          "post_date": "04/15/2022 10:30:13",
          "content": "<p>exactly, 24% of 28k images is public part</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1756088": "I've been training and testing my model with cropped dataset. I realize that the final test dataset won't be cropped. Should I add in my codes a function to crop the final test dataset before inferencing?",
    "1756109": "Usually yes, you should predict the same preprocessing (without some hard augmentation, or use it in TTA) like in training",
    "1756115": "Mind if I ask a question further into it? How am I supposed to figure out the final test dataset folder location  when I have no clue?",
    "1756232": "What do you mean under \"final test dataset\", you already have final test dataset, 28k images",
    "1756237": "I thought we're going to test another training set that consists of 88k images (28k:24% = x:76%) . Oh wait.... 76% means the other 76% of THE test dataset? So the current leaderboard scores are evaluated only on 24% of the test dataset result that we submit?",
    "1756238": "Oh my god. I was totally mistaken. Thank you so much!",
    "1756240": "exactly, 24% of 28k images is public part"
  },
  "source": "meta"
}