{
  "id": 416597,
  "title": "What's the difference between Individual annotations(human_individual_masks.npy) and aggregated ground truth annotations (human_pixel_masks.npy)?",
  "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/416597",
  "author_name": "Peter CXL",
  "post_date": "2023-06-12T09:26:38.307000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Does it mean that Individual annotations is training data and aggregated ground truth annotations is testing data? I'm not an English native and don't understant it well…<br>\nThose two words is from \"Dada\" page(<a href=\"https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)\" target=\"_blank\">https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)</a>.</p>",
  "messages": [
    {
      "id": 2296989,
      "postDate": "2023-06-12T09:47:41.317Z",
      "content": "<p>See the human_individual_masks.npy as a list of masks where each mask is what a single labeler saw from this image.<br>\nOn the other hand, the human_pixel_masks is the ground truth of this problem. they decided for each pixel of the 256x256 grid that if <strong>more</strong> than 50% of the labelers marked it as positive, it would be positive.</p>\n<p>Thats how it could have been made:</p>\n<pre><code>labels = np.load()\nintermediate_label = labels.mean(axis=)\nlabel = (intermediate_label &gt; ).astype(np.int32)\nnp.save(label, )\n</code></pre>",
      "rawMarkdown": "See the human_individual_masks.npy as a list of masks where each mask is what a single labeler saw from this image.\nOn the other hand, the human_pixel_masks is the ground truth of this problem. they decided for each pixel of the 256x256 grid that if **more** than 50% of the labelers marked it as positive, it would be positive.\n\nThats how it could have been made:\n```python\nlabels = np.load('human_individual_masks.npy')\nintermediate_label = labels.mean(axis=3)\nlabel = (intermediate_label > 0.5).astype(np.int32)\nnp.save(label, 'human_pixel_masks.npy')\n```\n",
      "votes": 3,
      "replies": [
        {
          "id": 2297370,
          "postDate": "2023-06-12T14:27:25.110Z",
          "content": "<p>Thx~ ^^ Janmpia</p>",
          "rawMarkdown": "Thx~ ^^ Janmpia",
          "votes": 1
        },
        {
          "id": 2366693,
          "postDate": "2023-07-31T06:42:20.870Z",
          "content": "<p>does that mean i dont really need to consider  human_individual_masks.npy while training ?</p>",
          "rawMarkdown": "does that mean i dont really need to consider  human_individual_masks.npy while training ?"
        }
      ]
    },
    {
      "id": 2296956,
      "postDate": "2023-06-12T09:26:38.307Z",
      "content": "<p>Does it mean that Individual annotations is training data and aggregated ground truth annotations is testing data? I'm not an English native and don't understant it well…<br>\nThose two words is from \"Dada\" page(<a href=\"https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)\" target=\"_blank\">https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)</a>.</p>",
      "rawMarkdown": "Does it mean that Individual annotations is training data and aggregated ground truth annotations is testing data? I'm not an English native and don't understant it well...\nThose two words is from \"Dada\" page(https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)."
    }
  ],
  "comments": [
    {
      "id": 2296989,
      "author_name": "JEANMPIA",
      "author_url": "",
      "post_date": "2023-06-12T09:47:41.317000",
      "content": "<p>See the human_individual_masks.npy as a list of masks where each mask is what a single labeler saw from this image.<br>\nOn the other hand, the human_pixel_masks is the ground truth of this problem. they decided for each pixel of the 256x256 grid that if <strong>more</strong> than 50% of the labelers marked it as positive, it would be positive.</p>\n<p>Thats how it could have been made:</p>\n<pre><code>labels = np.load()\nintermediate_label = labels.mean(axis=)\nlabel = (intermediate_label &gt; ).astype(np.int32)\nnp.save(label, )\n</code></pre>",
      "votes": 3,
      "replies": [
        {
          "id": 2297370,
          "author_name": "Peter CXL",
          "author_url": "",
          "post_date": "2023-06-12T14:27:25.110000",
          "content": "<p>Thx~ ^^ Janmpia</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 2366693,
          "author_name": "ADITHYA L BHAT",
          "author_url": "",
          "post_date": "2023-07-31T06:42:20.870000",
          "content": "<p>does that mean i dont really need to consider  human_individual_masks.npy while training ?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2296989": "See the human_individual_masks.npy as a list of masks where each mask is what a single labeler saw from this image.\nOn the other hand, the human_pixel_masks is the ground truth of this problem. they decided for each pixel of the 256x256 grid that if **more** than 50% of the labelers marked it as positive, it would be positive.\n\nThats how it could have been made:\n```python\nlabels = np.load('human_individual_masks.npy')\nintermediate_label = labels.mean(axis=3)\nlabel = (intermediate_label > 0.5).astype(np.int32)\nnp.save(label, 'human_pixel_masks.npy')\n```\n",
    "2296956": "Does it mean that Individual annotations is training data and aggregated ground truth annotations is testing data? I'm not an English native and don't understant it well...\nThose two words is from \"Dada\" page(https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming/data)."
  }
}