{
  "id": 400436,
  "title": "Full size PNG image dataset (37Gb -> 5Gb)",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/400436",
  "author_name": "",
  "post_date": "2023-04-08T11:21:02.790919500Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have created PNG version of the dataset</p>\n<p>each image was generated from below cv2 code</p>\n<pre><code> ():\n     (img - img.()) / (img.() - img.())\n\nimg = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)\nimg = (norm01(img) * ).astype(np.uint8)\ncv2.imwrite(save_path, img)\n</code></pre>\n<p>link: <a href=\"https://www.kaggle.com/datasets/ptran1203/vesuvius-png\" target=\"_blank\">https://www.kaggle.com/datasets/ptran1203/vesuvius-png</a></p>",
  "messages": [
    {
      "id": "2214327",
      "postDate": "04/08/2023 11:21:02",
      "content": "<p>I have created PNG version of the dataset</p>\n<p>each image was generated from below cv2 code</p>\n<pre><code> ():\n     (img - img.()) / (img.() - img.())\n\nimg = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)\nimg = (norm01(img) * ).astype(np.uint8)\ncv2.imwrite(save_path, img)\n</code></pre>\n<p>link: <a href=\"https://www.kaggle.com/datasets/ptran1203/vesuvius-png\" target=\"_blank\">https://www.kaggle.com/datasets/ptran1203/vesuvius-png</a></p>",
      "rawMarkdown": "I have created PNG version of the dataset\n\neach image was generated from below cv2 code\n\n```python\ndef norm01(img):\n    return (img - img.min()) / (img.max() - img.min())\n\nimg = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)\nimg = (norm01(img) * 255.0).astype(np.uint8)\ncv2.imwrite(save_path, img)\n```\n\nlink: https://www.kaggle.com/datasets/ptran1203/vesuvius-png",
      "votes": null
    },
    {
      "id": "2217815",
      "postDate": "04/11/2023 07:30:35",
      "content": "<p>Have you noticed any differences in model performance when using the PNG version of the dataset compared to the original?</p>",
      "rawMarkdown": "Have you noticed any differences in model performance when using the PNG version of the dataset compared to the original?",
      "votes": null
    },
    {
      "id": "2218326",
      "postDate": "04/11/2023 15:24:10",
      "content": "<p>I did not compare yet, But some information may lost when I convert to [0-255] uint8, original image value range is very large</p>",
      "rawMarkdown": "I did not compare yet, But some information may lost when I convert to [0-255] uint8, original image value range is very large",
      "votes": null
    },
    {
      "id": "2226308",
      "postDate": "04/18/2023 20:31:08",
      "content": "<p>Awesome, thanks!</p>",
      "rawMarkdown": "Awesome, thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2217815,
      "author_name": "yus002",
      "author_url": "",
      "post_date": "04/11/2023 07:30:35",
      "content": "<p>Have you noticed any differences in model performance when using the PNG version of the dataset compared to the original?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2218326,
          "author_name": "ptran1203",
          "author_url": "",
          "post_date": "04/11/2023 15:24:10",
          "content": "<p>I did not compare yet, But some information may lost when I convert to [0-255] uint8, original image value range is very large</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2226308,
      "author_name": "authman",
      "author_url": "",
      "post_date": "04/18/2023 20:31:08",
      "content": "<p>Awesome, thanks!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2214327": "I have created PNG version of the dataset\n\neach image was generated from below cv2 code\n\n```python\ndef norm01(img):\n    return (img - img.min()) / (img.max() - img.min())\n\nimg = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)\nimg = (norm01(img) * 255.0).astype(np.uint8)\ncv2.imwrite(save_path, img)\n```\n\nlink: https://www.kaggle.com/datasets/ptran1203/vesuvius-png",
    "2217815": "Have you noticed any differences in model performance when using the PNG version of the dataset compared to the original?",
    "2218326": "I did not compare yet, But some information may lost when I convert to [0-255] uint8, original image value range is very large",
    "2226308": "Awesome, thanks!"
  },
  "source": "meta"
}