{
  "id": 450292,
  "title": "Question about the Test Time Limitation",
  "url": "/competitions/UBC-OCEAN/discussion/450292",
  "author_name": "Zijie Fang",
  "post_date": "2023-10-23T20:15:21.046000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I have tested that simply loading the original <code>.png</code>s using the <code>cv2</code> module and detecting the background areas require over 12h when using the P100 GPU in Kaggle notebook for the training set. Although the test set is smaller, the 12h time limitation is still very strict, even for loading all the large-scaled <code>.png</code>s.</p>\n<p>Are there anyone having ideas about this problem? If we should only utilize the thumbnails to satisfy the test time limitation? (but the thumbnails lose a great number of morphology details, which may lead to performance degradation :/)</p>",
  "messages": [
    {
      "id": 2496203,
      "postDate": "2023-10-23T20:15:21.047Z",
      "content": "<p>I have tested that simply loading the original <code>.png</code>s using the <code>cv2</code> module and detecting the background areas require over 12h when using the P100 GPU in Kaggle notebook for the training set. Although the test set is smaller, the 12h time limitation is still very strict, even for loading all the large-scaled <code>.png</code>s.</p>\n<p>Are there anyone having ideas about this problem? If we should only utilize the thumbnails to satisfy the test time limitation? (but the thumbnails lose a great number of morphology details, which may lead to performance degradation :/)</p>",
      "rawMarkdown": "I have tested that simply loading the original `.png`s using the `cv2` module and detecting the background areas require over 12h when using the P100 GPU in Kaggle notebook for the training set. Although the test set is smaller, the 12h time limitation is still very strict, even for loading all the large-scaled `.png`s.\n\nAre there anyone having ideas about this problem? If we should only utilize the thumbnails to satisfy the test time limitation? (but the thumbnails lose a great number of morphology details, which may lead to performance degradation :/)",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2496203": "I have tested that simply loading the original `.png`s using the `cv2` module and detecting the background areas require over 12h when using the P100 GPU in Kaggle notebook for the training set. Although the test set is smaller, the 12h time limitation is still very strict, even for loading all the large-scaled `.png`s.\n\nAre there anyone having ideas about this problem? If we should only utilize the thumbnails to satisfy the test time limitation? (but the thumbnails lose a great number of morphology details, which may lead to performance degradation :/)"
  }
}