{
  "id": 263921,
  "title": "A small suggestion for solution sharing",
  "url": "/competitions/siim-covid19-detection/discussion/263921",
  "author_name": "Xin Yi",
  "post_date": "2021-08-10T15:21:31.080000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Congratulations to all the winners and the openness of solution sharing. This is truly a very interesting contest. I have seen a lot of interesting ideas which I have never thought of or have neglected the whole time. For example, adding a fifth none class in the 4 class classification model, using the whole image as bounding boxes for detection and pretrain the model with the other chest X-ray datasets.</p>\n<p>However, there are mainly two things that I wish could be presented in the solutions. The first is the type of compute infrastructure (gpu, tpu, what type) people are using and the second is the amount of effort (hours) people have devoted into the competition. I know the two factors are not independent. People with high level of expertise probably won’t need as much time of exploration as the junior people. But I believe people can still use it to calibrate themselves (or help better manage resources) and appreciate how much dedication and creativity it gonna be required to have a good stand in the next contest.</p>\n<p>Thank you all!</p>",
  "messages": [
    {
      "id": 1464452,
      "postDate": "2021-08-10T15:21:31.080Z",
      "content": "<p>Congratulations to all the winners and the openness of solution sharing. This is truly a very interesting contest. I have seen a lot of interesting ideas which I have never thought of or have neglected the whole time. For example, adding a fifth none class in the 4 class classification model, using the whole image as bounding boxes for detection and pretrain the model with the other chest X-ray datasets.</p>\n<p>However, there are mainly two things that I wish could be presented in the solutions. The first is the type of compute infrastructure (gpu, tpu, what type) people are using and the second is the amount of effort (hours) people have devoted into the competition. I know the two factors are not independent. People with high level of expertise probably won’t need as much time of exploration as the junior people. But I believe people can still use it to calibrate themselves (or help better manage resources) and appreciate how much dedication and creativity it gonna be required to have a good stand in the next contest.</p>\n<p>Thank you all!</p>",
      "rawMarkdown": "Congratulations to all the winners and the openness of solution sharing. This is truly a very interesting contest. I have seen a lot of interesting ideas which I have never thought of or have neglected the whole time. For example, adding a fifth none class in the 4 class classification model, using the whole image as bounding boxes for detection and pretrain the model with the other chest X-ray datasets.\n\nHowever, there are mainly two things that I wish could be presented in the solutions. The first is the type of compute infrastructure (gpu, tpu, what type) people are using and the second is the amount of effort (hours) people have devoted into the competition. I know the two factors are not independent. People with high level of expertise probably won’t need as much time of exploration as the junior people. But I believe people can still use it to calibrate themselves (or help better manage resources) and appreciate how much dedication and creativity it gonna be required to have a good stand in the next contest.\n\nThank you all!",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1464452": "Congratulations to all the winners and the openness of solution sharing. This is truly a very interesting contest. I have seen a lot of interesting ideas which I have never thought of or have neglected the whole time. For example, adding a fifth none class in the 4 class classification model, using the whole image as bounding boxes for detection and pretrain the model with the other chest X-ray datasets.\n\nHowever, there are mainly two things that I wish could be presented in the solutions. The first is the type of compute infrastructure (gpu, tpu, what type) people are using and the second is the amount of effort (hours) people have devoted into the competition. I know the two factors are not independent. People with high level of expertise probably won’t need as much time of exploration as the junior people. But I believe people can still use it to calibrate themselves (or help better manage resources) and appreciate how much dedication and creativity it gonna be required to have a good stand in the next contest.\n\nThank you all!"
  }
}