{
  "id": 416165,
  "title": "How can we produce good models, with such low resource environment",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/416165",
  "author_name": "",
  "post_date": "2023-06-09T23:53:34.258814100Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Yep, I know, you will say upgrade the notebook.  I tried, but GCP doesn't seem to have any GPU instances.  </p>\n<p>So I am getting an out of memory error now, </p>\n<p>What do people do to get over this obstacle</p>",
  "messages": [
    {
      "id": "2294279",
      "postDate": "06/09/2023 23:53:34",
      "content": "<p>Yep, I know, you will say upgrade the notebook.  I tried, but GCP doesn't seem to have any GPU instances.  </p>\n<p>So I am getting an out of memory error now, </p>\n<p>What do people do to get over this obstacle</p>",
      "rawMarkdown": "Yep, I know, you will say upgrade the notebook.  I tried, but GCP doesn't seem to have any GPU instances.  \n\nSo I am getting an out of memory error now, \n\nWhat do people do to get over this obstacle",
      "votes": null
    },
    {
      "id": "2294545",
      "postDate": "06/10/2023 06:32:33",
      "content": "<p>I suspect we are all struggling on this one. </p>\n<p>Currently resort to performing inference on CPU, go 2D, etc.</p>",
      "rawMarkdown": "I suspect we are all struggling on this one. \n\nCurrently resort to performing inference on CPU, go 2D, etc.",
      "votes": null
    },
    {
      "id": "2295463",
      "postDate": "06/11/2023 01:52:32",
      "content": "<p>Looks like a repeat of the 2020 Fake Video competitions.  Low end systems caused people to make work arounds that resulted in weak models.  People will work with what you give them</p>\n<p>I have been waiting on the availability of a single GPU on GCP for 4 years now. Kaggle people, help us please!</p>",
      "rawMarkdown": "Looks like a repeat of the 2020 Fake Video competitions.  Low end systems caused people to make work arounds that resulted in weak models.  People will work with what you give them\n\nI have been waiting on the availability of a single GPU on GCP for 4 years now. Kaggle people, help us please!",
      "votes": null
    },
    {
      "id": "2295516",
      "postDate": "06/11/2023 03:59:07",
      "content": "<p>check my post.</p>\n<p>if you have low resource just use encoder and output prediction mask downsized at 1/32.<br>\nThis already gives you lb score in range of 0.7x </p>\n<p>you actually do not need decoder.</p>",
      "rawMarkdown": "check my post.\n\nif you have low resource just use encoder and output prediction mask downsized at 1/32.\nThis already gives you lb score in range of 0.7x \n\nyou actually do not need decoder.",
      "votes": null
    },
    {
      "id": "2301010",
      "postDate": "06/13/2023 15:07:29",
      "content": "<p>I can't see your post.</p>",
      "rawMarkdown": "I can't see your post.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2294545,
      "author_name": "vcolliym",
      "author_url": "",
      "post_date": "06/10/2023 06:32:33",
      "content": "<p>I suspect we are all struggling on this one. </p>\n<p>Currently resort to performing inference on CPU, go 2D, etc.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2295463,
          "author_name": "abualabed",
          "author_url": "",
          "post_date": "06/11/2023 01:52:32",
          "content": "<p>Looks like a repeat of the 2020 Fake Video competitions.  Low end systems caused people to make work arounds that resulted in weak models.  People will work with what you give them</p>\n<p>I have been waiting on the availability of a single GPU on GCP for 4 years now. Kaggle people, help us please!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2295516,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "06/11/2023 03:59:07",
      "content": "<p>check my post.</p>\n<p>if you have low resource just use encoder and output prediction mask downsized at 1/32.<br>\nThis already gives you lb score in range of 0.7x </p>\n<p>you actually do not need decoder.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2301010,
          "author_name": "billmorrisson",
          "author_url": "",
          "post_date": "06/13/2023 15:07:29",
          "content": "<p>I can't see your post.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2294279": "Yep, I know, you will say upgrade the notebook.  I tried, but GCP doesn't seem to have any GPU instances.  \n\nSo I am getting an out of memory error now, \n\nWhat do people do to get over this obstacle",
    "2294545": "I suspect we are all struggling on this one. \n\nCurrently resort to performing inference on CPU, go 2D, etc.",
    "2295463": "Looks like a repeat of the 2020 Fake Video competitions.  Low end systems caused people to make work arounds that resulted in weak models.  People will work with what you give them\n\nI have been waiting on the availability of a single GPU on GCP for 4 years now. Kaggle people, help us please!",
    "2295516": "check my post.\n\nif you have low resource just use encoder and output prediction mask downsized at 1/32.\nThis already gives you lb score in range of 0.7x \n\nyou actually do not need decoder.",
    "2301010": "I can't see your post."
  },
  "source": "meta"
}