{
  "id": 160485,
  "title": "Low Public Score ",
  "url": "/competitions/global-wheat-detection/discussion/160485",
  "author_name": "",
  "post_date": "2020-06-21T12:22:53.070518900Z",
  "votes": 1,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Hi, I'm a begginer and I just submit my notebook, and I don't understand why the public score is so low (0.0009). My prediction are good, not best fit but pretty good. I would really appriciate feedback to my notebook:\n<a href=\"https://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413\">https://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413</a></p>\n\n<p>Elad</p>",
  "messages": [
    {
      "id": "895512",
      "postDate": "06/21/2020 12:22:53",
      "content": "<p>Hi, I'm a begginer and I just submit my notebook, and I don't understand why the public score is so low (0.0009). My prediction are good, not best fit but pretty good. I would really appriciate feedback to my notebook:\n<a href=\"https://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413\">https://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413</a></p>\n\n<p>Elad</p>",
      "rawMarkdown": "Hi, I'm a begginer and I just submit my notebook, and I don't understand why the public score is so low (0.0009). My prediction are good, not best fit but pretty good. I would really appriciate feedback to my notebook:\nhttps://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413\n\nElad",
      "votes": null
    },
    {
      "id": "895569",
      "postDate": "06/21/2020 13:09:00",
      "content": "<p>The mistake is here: </p>\n\n<blockquote>\n  <p>submit = pd.read_csv(testFile)\n  for idx,imgid in enumerate(submit['image_id']):\n      # Attch the prediction string into a submitted file:\n      submit.iloc[idx,1]= report[report['image_id']==imgid].values[0,1]</p>\n</blockquote>\n\n<p>You're only prediciting bboxes of images in 'sample_submission'. However these iamges are a small part of the test folder. So during the submission you should predict bboxes images in the test folder.</p>\n\n<p>Good Luck!</p>",
      "rawMarkdown": "The mistake is here: \n\n&gt; submit = pd.read_csv(testFile)\nfor idx,imgid in enumerate(submit['image_id']):\n    # Attch the prediction string into a submitted file:\n    submit.iloc[idx,1]= report[report['image_id']==imgid].values[0,1]\n\nYou're only prediciting bboxes of images in 'sample_submission'. However these iamges are a small part of the test folder. So during the submission you should predict bboxes images in the test folder.\n\nGood Luck!",
      "votes": null
    },
    {
      "id": "896435",
      "postDate": "06/22/2020 07:12:40",
      "content": "<p>But the 'sample_submission' are exactly the same amount of images as the test folder.\nI don't understand what should I replace on the code :(</p>",
      "rawMarkdown": "But the 'sample_submission' are exactly the same amount of images as the test folder.\nI don't understand what should I replace on the code :(",
      "votes": null
    },
    {
      "id": "896756",
      "postDate": "06/22/2020 12:17:18",
      "content": "<p>The test folder you see is just to give you an idea of the format. The actual test set is around 1000 images. Look at existing public notebooks to get an idea how to perform the inference..</p>",
      "rawMarkdown": "The test folder you see is just to give you an idea of the format. The actual test set is around 1000 images. Look at existing public notebooks to get an idea how to perform the inference..",
      "votes": null
    },
    {
      "id": "897289",
      "postDate": "06/22/2020 18:54:51",
      "content": "<p>Hey,\nI tried to change the submission format:\n <a href=\"https://www.kaggle.com/eladsoll/gwd-es\">https://www.kaggle.com/eladsoll/gwd-es</a>\nBut I got failed score. what should I do? guide please</p>",
      "rawMarkdown": "Hey,\nI tried to change the submission format:\n https://www.kaggle.com/eladsoll/gwd-es\nBut I got failed score. what should I do? guide please",
      "votes": null
    },
    {
      "id": "897839",
      "postDate": "06/23/2020 06:12:20",
      "content": "<p>Did you check public notebook such as <a href=\"https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\">https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train</a> ? You can look at his inference notebook and start from here.</p>",
      "rawMarkdown": "Did you check public notebook such as https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train ? You can look at his inference notebook and start from here.",
      "votes": null
    },
    {
      "id": "898036",
      "postDate": "06/23/2020 08:37:40",
      "content": "<p>Hi Etienne I really can't see differences between my notebook and others.\nHere is a simillar notbook to my:\n<a href=\"https://www.kaggle.com/aviralpamecha/inference-keras-retinanet\">https://www.kaggle.com/aviralpamecha/inference-keras-retinanet</a>\nIt's my first challenge on kaggle , I want to learn where is my mistake , I will really appriciate help/guide, what to do. It's not clear to me what else to do.\n<strong>Thank you all for the fast response and your patience</strong></p>",
      "rawMarkdown": "Hi Etienne I really can't see differences between my notebook and others.\nHere is a simillar notbook to my:\nhttps://www.kaggle.com/aviralpamecha/inference-keras-retinanet\nIt's my first challenge on kaggle , I want to learn where is my mistake , I will really appriciate help/guide, what to do. It's not clear to me what else to do.\n**Thank you all for the fast response and your patience**",
      "votes": null
    },
    {
      "id": "899253",
      "postDate": "06/24/2020 05:35:54",
      "content": "<p>Maybe your model predict wheat head with a confidence lower than 0.5 ? the test set is quite different from the training dataset, and in your notebook you already have a low confidence for your object detection</p>",
      "rawMarkdown": "Maybe your model predict wheat head with a confidence lower than 0.5 ? the test set is quite different from the training dataset, and in your notebook you already have a low confidence for your object detection",
      "votes": null
    },
    {
      "id": "899834",
      "postDate": "06/24/2020 13:21:32",
      "content": "<p>hi,  you can check your input image size in your model \nin the inference , the box must be matching the image with size=1024,so you must use a alpha number by the original image size,  let it be 1024.     </p>",
      "rawMarkdown": "hi,  you can check your input image size in your model \nin the inference , the box must be matching the image with size=1024,so you must use a alpha number by the original image size,  let it be 1024.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 895569,
      "author_name": "ffares",
      "author_url": "",
      "post_date": "06/21/2020 13:09:00",
      "content": "<p>The mistake is here: </p>\n\n<blockquote>\n  <p>submit = pd.read_csv(testFile)\n  for idx,imgid in enumerate(submit['image_id']):\n      # Attch the prediction string into a submitted file:\n      submit.iloc[idx,1]= report[report['image_id']==imgid].values[0,1]</p>\n</blockquote>\n\n<p>You're only prediciting bboxes of images in 'sample_submission'. However these iamges are a small part of the test folder. So during the submission you should predict bboxes images in the test folder.</p>\n\n<p>Good Luck!</p>",
      "votes": null,
      "replies": [
        {
          "id": 896435,
          "author_name": "eladsoll",
          "author_url": "",
          "post_date": "06/22/2020 07:12:40",
          "content": "<p>But the 'sample_submission' are exactly the same amount of images as the test folder.\nI don't understand what should I replace on the code :(</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 896756,
          "author_name": "matthewmasters",
          "author_url": "",
          "post_date": "06/22/2020 12:17:18",
          "content": "<p>The test folder you see is just to give you an idea of the format. The actual test set is around 1000 images. Look at existing public notebooks to get an idea how to perform the inference..</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 897289,
          "author_name": "eladsoll",
          "author_url": "",
          "post_date": "06/22/2020 18:54:51",
          "content": "<p>Hey,\nI tried to change the submission format:\n <a href=\"https://www.kaggle.com/eladsoll/gwd-es\">https://www.kaggle.com/eladsoll/gwd-es</a>\nBut I got failed score. what should I do? guide please</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 897839,
          "author_name": "bendvd",
          "author_url": "",
          "post_date": "06/23/2020 06:12:20",
          "content": "<p>Did you check public notebook such as <a href=\"https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\">https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train</a> ? You can look at his inference notebook and start from here.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 898036,
          "author_name": "eladsoll",
          "author_url": "",
          "post_date": "06/23/2020 08:37:40",
          "content": "<p>Hi Etienne I really can't see differences between my notebook and others.\nHere is a simillar notbook to my:\n<a href=\"https://www.kaggle.com/aviralpamecha/inference-keras-retinanet\">https://www.kaggle.com/aviralpamecha/inference-keras-retinanet</a>\nIt's my first challenge on kaggle , I want to learn where is my mistake , I will really appriciate help/guide, what to do. It's not clear to me what else to do.\n<strong>Thank you all for the fast response and your patience</strong></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 899253,
      "author_name": "bendvd",
      "author_url": "",
      "post_date": "06/24/2020 05:35:54",
      "content": "<p>Maybe your model predict wheat head with a confidence lower than 0.5 ? the test set is quite different from the training dataset, and in your notebook you already have a low confidence for your object detection</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 899834,
      "author_name": "okbenggg",
      "author_url": "",
      "post_date": "06/24/2020 13:21:32",
      "content": "<p>hi,  you can check your input image size in your model \nin the inference , the box must be matching the image with size=1024,so you must use a alpha number by the original image size,  let it be 1024.     </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "895512": "Hi, I'm a begginer and I just submit my notebook, and I don't understand why the public score is so low (0.0009). My prediction are good, not best fit but pretty good. I would really appriciate feedback to my notebook:\nhttps://www.kaggle.com/eladsoll/gwd-es?scriptVersionId=37026413\n\nElad",
    "895569": "The mistake is here: \n\n&gt; submit = pd.read_csv(testFile)\nfor idx,imgid in enumerate(submit['image_id']):\n    # Attch the prediction string into a submitted file:\n    submit.iloc[idx,1]= report[report['image_id']==imgid].values[0,1]\n\nYou're only prediciting bboxes of images in 'sample_submission'. However these iamges are a small part of the test folder. So during the submission you should predict bboxes images in the test folder.\n\nGood Luck!",
    "896435": "But the 'sample_submission' are exactly the same amount of images as the test folder.\nI don't understand what should I replace on the code :(",
    "896756": "The test folder you see is just to give you an idea of the format. The actual test set is around 1000 images. Look at existing public notebooks to get an idea how to perform the inference..",
    "897289": "Hey,\nI tried to change the submission format:\n https://www.kaggle.com/eladsoll/gwd-es\nBut I got failed score. what should I do? guide please",
    "897839": "Did you check public notebook such as https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train ? You can look at his inference notebook and start from here.",
    "898036": "Hi Etienne I really can't see differences between my notebook and others.\nHere is a simillar notbook to my:\nhttps://www.kaggle.com/aviralpamecha/inference-keras-retinanet\nIt's my first challenge on kaggle , I want to learn where is my mistake , I will really appriciate help/guide, what to do. It's not clear to me what else to do.\n**Thank you all for the fast response and your patience**",
    "899253": "Maybe your model predict wheat head with a confidence lower than 0.5 ? the test set is quite different from the training dataset, and in your notebook you already have a low confidence for your object detection",
    "899834": "hi,  you can check your input image size in your model \nin the inference , the box must be matching the image with size=1024,so you must use a alpha number by the original image size,  let it be 1024."
  },
  "source": "meta"
}