{
  "id": 412117,
  "title": "ENSEMBLE-Approach",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/412117",
  "author_name": "Vishak K Bhat",
  "post_date": "2023-05-22T10:28:53.459000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<ul>\n<li><p><strong>Convert run-length encoding to pixel-wise masks:</strong> Convert the run-length encoding representations of each model's predictions into pixel-wise masks. This involves mapping the starting pixel and the number of consecutive pixels with ink to their corresponding positions in the mask. The result is a binary mask where the pixels corresponding to the object or class are set to 1, and the background pixels are set to 0.</p></li>\n<li><p><strong>Average the pixel-wise masks:</strong> Compute the element-wise average of the pixel-wise masks generated by each model. For each pixel position, sum the corresponding values from the masks and divide by the total number of models. This averaging process generates an averaged mask with pixel values between 0 and 1.</p></li>\n<li><p><strong>Apply a threshold:</strong> Apply a threshold of 0.5 to the averaged mask. Set any pixel value greater than 0.5 to 1, indicating the presence of the ink, and set values below or equal to 0.5 to 0, indicating the background.</p></li>\n<li><p><strong>Convert the thresholded mask to run-length encoding:</strong> Convert the thresholded mask back to run-length encoding format. Iterate over the pixels in the mask, identifying consecutive regions of 1s and encoding them as starting pixel positions and the number of consecutive pixels with ink. This step transforms the thresholded mask into a concise run-length encoding representation.</p></li>\n</ul>",
  "messages": [
    {
      "id": 2269246,
      "postDate": "2023-05-22T10:28:53.460Z",
      "content": "<ul>\n<li><p><strong>Convert run-length encoding to pixel-wise masks:</strong> Convert the run-length encoding representations of each model's predictions into pixel-wise masks. This involves mapping the starting pixel and the number of consecutive pixels with ink to their corresponding positions in the mask. The result is a binary mask where the pixels corresponding to the object or class are set to 1, and the background pixels are set to 0.</p></li>\n<li><p><strong>Average the pixel-wise masks:</strong> Compute the element-wise average of the pixel-wise masks generated by each model. For each pixel position, sum the corresponding values from the masks and divide by the total number of models. This averaging process generates an averaged mask with pixel values between 0 and 1.</p></li>\n<li><p><strong>Apply a threshold:</strong> Apply a threshold of 0.5 to the averaged mask. Set any pixel value greater than 0.5 to 1, indicating the presence of the ink, and set values below or equal to 0.5 to 0, indicating the background.</p></li>\n<li><p><strong>Convert the thresholded mask to run-length encoding:</strong> Convert the thresholded mask back to run-length encoding format. Iterate over the pixels in the mask, identifying consecutive regions of 1s and encoding them as starting pixel positions and the number of consecutive pixels with ink. This step transforms the thresholded mask into a concise run-length encoding representation.</p></li>\n</ul>",
      "rawMarkdown": "- **Convert run-length encoding to pixel-wise masks:** Convert the run-length encoding representations of each model's predictions into pixel-wise masks. This involves mapping the starting pixel and the number of consecutive pixels with ink to their corresponding positions in the mask. The result is a binary mask where the pixels corresponding to the object or class are set to 1, and the background pixels are set to 0.\n\n- **Average the pixel-wise masks:** Compute the element-wise average of the pixel-wise masks generated by each model. For each pixel position, sum the corresponding values from the masks and divide by the total number of models. This averaging process generates an averaged mask with pixel values between 0 and 1.\n\n- **Apply a threshold:** Apply a threshold of 0.5 to the averaged mask. Set any pixel value greater than 0.5 to 1, indicating the presence of the ink, and set values below or equal to 0.5 to 0, indicating the background.\n\n- **Convert the thresholded mask to run-length encoding:** Convert the thresholded mask back to run-length encoding format. Iterate over the pixels in the mask, identifying consecutive regions of 1s and encoding them as starting pixel positions and the number of consecutive pixels with ink. This step transforms the thresholded mask into a concise run-length encoding representation.\n\n",
      "votes": 2
    },
    {
      "id": 2270213,
      "postDate": "2023-05-23T04:58:07.207Z",
      "content": "<p>Nice work Vishak!!!!</p>",
      "rawMarkdown": "Nice work Vishak!!!!",
      "votes": 1
    },
    {
      "id": 2270558,
      "postDate": "2023-05-23T08:23:17.747Z",
      "content": "<p>Nice work! Thanks for sharing the ensemble idea, How many model have you ensemble?</p>",
      "rawMarkdown": "Nice work! Thanks for sharing the ensemble idea, How many model have you ensemble?"
    },
    {
      "id": 2270555,
      "postDate": "2023-05-23T08:22:22.547Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2270213,
      "author_name": "Arushika Bansal",
      "author_url": "",
      "post_date": "2023-05-23T04:58:07.207000",
      "content": "<p>Nice work Vishak!!!!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2270558,
      "author_name": "Dewei Chen",
      "author_url": "",
      "post_date": "2023-05-23T08:23:17.747000",
      "content": "<p>Nice work! Thanks for sharing the ensemble idea, How many model have you ensemble?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2270555,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-05-23T08:22:22.547000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2269246": "- **Convert run-length encoding to pixel-wise masks:** Convert the run-length encoding representations of each model's predictions into pixel-wise masks. This involves mapping the starting pixel and the number of consecutive pixels with ink to their corresponding positions in the mask. The result is a binary mask where the pixels corresponding to the object or class are set to 1, and the background pixels are set to 0.\n\n- **Average the pixel-wise masks:** Compute the element-wise average of the pixel-wise masks generated by each model. For each pixel position, sum the corresponding values from the masks and divide by the total number of models. This averaging process generates an averaged mask with pixel values between 0 and 1.\n\n- **Apply a threshold:** Apply a threshold of 0.5 to the averaged mask. Set any pixel value greater than 0.5 to 1, indicating the presence of the ink, and set values below or equal to 0.5 to 0, indicating the background.\n\n- **Convert the thresholded mask to run-length encoding:** Convert the thresholded mask back to run-length encoding format. Iterate over the pixels in the mask, identifying consecutive regions of 1s and encoding them as starting pixel positions and the number of consecutive pixels with ink. This step transforms the thresholded mask into a concise run-length encoding representation.\n\n",
    "2270213": "Nice work Vishak!!!!",
    "2270558": "Nice work! Thanks for sharing the ensemble idea, How many model have you ensemble?",
    "2270555": ""
  }
}