{
  "id": 378042,
  "title": "Inference time",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/378042",
  "author_name": "Zubs",
  "post_date": "2023-01-14T07:22:51.915000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>Given the max time limit is 9 hours for ~8000 patients, that comes out to 4 seconds per patient with 2-4 images. Does that sound too tight or is it par for the course? Does that mean heavy models like transformers are out of the question?</p>\n<p>Thanks.</p>",
  "messages": [
    {
      "id": 2099289,
      "postDate": "2023-01-14T10:12:45.167Z",
      "content": "<p>If you use GPU for converting DCM files to PNG/JPG (for example, using DALI), it takes about 2.5 ~ 3.5 hours for me (depending on the pre-processing steps and algorithms applied to each and every DCM file). So, I would have at least 5 hours to perform inference on the hidden test set with roughly 34K images. I believe that 5 hours shall be enough to run some transformers based model inference.</p>",
      "rawMarkdown": "If you use GPU for converting DCM files to PNG/JPG (for example, using DALI), it takes about 2.5 ~ 3.5 hours for me (depending on the pre-processing steps and algorithms applied to each and every DCM file). So, I would have at least 5 hours to perform inference on the hidden test set with roughly 34K images. I believe that 5 hours shall be enough to run some transformers based model inference.",
      "votes": 2,
      "replies": [
        {
          "id": 2100556,
          "postDate": "2023-01-15T08:59:05.353Z",
          "content": "<p>Thanks for the input!</p>",
          "rawMarkdown": "Thanks for the input!"
        }
      ]
    },
    {
      "id": 2099113,
      "postDate": "2023-01-14T07:22:51.917Z",
      "content": "<p>Hi,</p>\n<p>Given the max time limit is 9 hours for ~8000 patients, that comes out to 4 seconds per patient with 2-4 images. Does that sound too tight or is it par for the course? Does that mean heavy models like transformers are out of the question?</p>\n<p>Thanks.</p>",
      "rawMarkdown": "Hi,\n\nGiven the max time limit is 9 hours for ~8000 patients, that comes out to 4 seconds per patient with 2-4 images. Does that sound too tight or is it par for the course? Does that mean heavy models like transformers are out of the question?\n\nThanks.",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 2099289,
      "author_name": "Rasoul Mojtahedzadeh",
      "author_url": "",
      "post_date": "2023-01-14T10:12:45.167000",
      "content": "<p>If you use GPU for converting DCM files to PNG/JPG (for example, using DALI), it takes about 2.5 ~ 3.5 hours for me (depending on the pre-processing steps and algorithms applied to each and every DCM file). So, I would have at least 5 hours to perform inference on the hidden test set with roughly 34K images. I believe that 5 hours shall be enough to run some transformers based model inference.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 2100556,
          "author_name": "Zubs",
          "author_url": "",
          "post_date": "2023-01-15T08:59:05.353000",
          "content": "<p>Thanks for the input!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2099289": "If you use GPU for converting DCM files to PNG/JPG (for example, using DALI), it takes about 2.5 ~ 3.5 hours for me (depending on the pre-processing steps and algorithms applied to each and every DCM file). So, I would have at least 5 hours to perform inference on the hidden test set with roughly 34K images. I believe that 5 hours shall be enough to run some transformers based model inference.",
    "2099113": "Hi,\n\nGiven the max time limit is 9 hours for ~8000 patients, that comes out to 4 seconds per patient with 2-4 images. Does that sound too tight or is it par for the course? Does that mean heavy models like transformers are out of the question?\n\nThanks."
  }
}