{
  "id": 32690,
  "title": "Clarification about train data modification",
  "url": "/competitions/noaa-fisheries-steller-sea-lion-population-count/discussion/32690",
  "author_name": "",
  "post_date": "2017-05-08T13:33:26.340894100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have posted this question originally in this thread: <a href=\"https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895\">https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895</a> \nbut I think it should be separate thread and I'm looking for official answer from competition organizers.</p>\n\n<ol>\n<li>Can we train directly on TrainDotted? i.e. inpaint color dots and train on these images? as I understand for now you just suggest to drop 'bad' images?</li>\n<li>Reported number of species is equal to number of dots in TrainDotted image (not the 'real' number of sea lions on Train image)?</li>\n<li>Can we manually edit number of dots and their location in train data? (for example if point is not on center of object or if we suspect that object should be another class)</li>\n<li>Can we manually remove images from train data? can we build patch based training data by hand(for example exclude patches with birds)? </li>\n</ol>",
  "messages": [
    {
      "id": "181078",
      "postDate": "05/08/2017 13:33:26",
      "content": "<p>I have posted this question originally in this thread: <a href=\"https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895\">https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895</a> \nbut I think it should be separate thread and I'm looking for official answer from competition organizers.</p>\n\n<ol>\n<li>Can we train directly on TrainDotted? i.e. inpaint color dots and train on these images? as I understand for now you just suggest to drop 'bad' images?</li>\n<li>Reported number of species is equal to number of dots in TrainDotted image (not the 'real' number of sea lions on Train image)?</li>\n<li>Can we manually edit number of dots and their location in train data? (for example if point is not on center of object or if we suspect that object should be another class)</li>\n<li>Can we manually remove images from train data? can we build patch based training data by hand(for example exclude patches with birds)? </li>\n</ol>",
      "rawMarkdown": "I have posted this question originally in this thread: https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895 \nbut I think it should be separate thread and I'm looking for official answer from competition organizers.\n\n1. Can we train directly on TrainDotted? i.e. inpaint color dots and train on these images? as I understand for now you just suggest to drop 'bad' images?\n2. Reported number of species is equal to number of dots in TrainDotted image (not the 'real' number of sea lions on Train image)?\n3. Can we manually edit number of dots and their location in train data? (for example if point is not on center of object or if we suspect that object should be another class)\n4. Can we manually remove images from train data? can we build patch based training data by hand(for example exclude patches with birds)?",
      "votes": null
    },
    {
      "id": "181144",
      "postDate": "05/08/2017 16:29:24",
      "content": "<p>I think that unless otherwise noted, you can do whatever you want with provided train data: use only a subset, modify it (e.g. correcting mistakes), add additional labels (e.g. add bounding boxes), etc.</p>",
      "rawMarkdown": "I think that unless otherwise noted, you can do whatever you want with provided train data: use only a subset, modify it (e.g. correcting mistakes), add additional labels (e.g. add bounding boxes), etc.",
      "votes": null
    },
    {
      "id": "181638",
      "postDate": "05/10/2017 11:13:22",
      "content": "<p>Question 3 have already been answered by Admin (DataCanary).<br>\nDataCanary said manual annotations/cropping of the training data are allowed.<br>\n<a href=\"https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502\">https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502</a></p>",
      "rawMarkdown": "Question 3 have already been answered by Admin (DataCanary).<br>\nDataCanary said manual annotations/cropping of the training data are allowed.<br>\nhttps://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 181144,
      "author_name": "lopuhin",
      "author_url": "",
      "post_date": "05/08/2017 16:29:24",
      "content": "<p>I think that unless otherwise noted, you can do whatever you want with provided train data: use only a subset, modify it (e.g. correcting mistakes), add additional labels (e.g. add bounding boxes), etc.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 181638,
      "author_name": "toshik",
      "author_url": "",
      "post_date": "05/10/2017 11:13:22",
      "content": "<p>Question 3 have already been answered by Admin (DataCanary).<br>\nDataCanary said manual annotations/cropping of the training data are allowed.<br>\n<a href=\"https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502\">https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "181078": "I have posted this question originally in this thread: https://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/30895 \nbut I think it should be separate thread and I'm looking for official answer from competition organizers.\n\n1. Can we train directly on TrainDotted? i.e. inpaint color dots and train on these images? as I understand for now you just suggest to drop 'bad' images?\n2. Reported number of species is equal to number of dots in TrainDotted image (not the 'real' number of sea lions on Train image)?\n3. Can we manually edit number of dots and their location in train data? (for example if point is not on center of object or if we suspect that object should be another class)\n4. Can we manually remove images from train data? can we build patch based training data by hand(for example exclude patches with birds)?",
    "181144": "I think that unless otherwise noted, you can do whatever you want with provided train data: use only a subset, modify it (e.g. correcting mistakes), add additional labels (e.g. add bounding boxes), etc.",
    "181638": "Question 3 have already been answered by Admin (DataCanary).<br>\nDataCanary said manual annotations/cropping of the training data are allowed.<br>\nhttps://www.kaggle.com/c/noaa-fisheries-steller-sea-lion-population-count/discussion/31048#172502"
  },
  "source": "meta"
}