{
  "id": 301703,
  "title": "what resolution did you use？",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/301703",
  "author_name": "",
  "post_date": "2022-01-19T03:37:41.074706300Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi，every, resolution 1280 in this competition seems to get 0.4+ LB score, if double of it, 0.5+ will be got. So, what resolution did you use？</p>",
  "messages": [
    {
      "id": "1656016",
      "postDate": "01/19/2022 03:37:41",
      "content": "<p>Hi，every, resolution 1280 in this competition seems to get 0.4+ LB score, if double of it, 0.5+ will be got. So, what resolution did you use？</p>",
      "rawMarkdown": "Hi，every, resolution 1280 in this competition seems to get 0.4+ LB score, if double of it, 0.5+ will be got. So, what resolution did you use？",
      "votes": null
    },
    {
      "id": "1656714",
      "postDate": "01/19/2022 15:05:48",
      "content": "<p>It depends on scenarios.  if you read other hotness discussion, you would find that in crease size to  around 2x 2.5x , pretty good LB.    in other situations, maybe reduce image size.</p>",
      "rawMarkdown": "It depends on scenarios.  if you read other hotness discussion, you would find that in crease size to  around 2x 2.5x , pretty good LB.    in other situations, maybe reduce image size.",
      "votes": null
    },
    {
      "id": "1660004",
      "postDate": "01/22/2022 09:56:48",
      "content": "<p>HI coallar, I ended up using a resolution of 3072x3072, it was the largest I could go with my GPU (1080TI). I unfortunately don't have a 3090 <a href=\"https://www.kaggle.com/steamedsheep/yolov5-high-resolution-training\" target=\"_blank\">like sheep</a> 😂</p>",
      "rawMarkdown": "HI coallar, I ended up using a resolution of 3072x3072, it was the largest I could go with my GPU (1080TI). I unfortunately don't have a 3090 [like sheep](https://www.kaggle.com/steamedsheep/yolov5-high-resolution-training) 😂",
      "votes": null
    },
    {
      "id": "1660011",
      "postDate": "01/22/2022 10:10:18",
      "content": "<p>my team partner uses 2080Ti, the resolution of 3600 was used, which LB got 0.57+</p>",
      "rawMarkdown": "my team partner uses 2080Ti, the resolution of 3600 was used, which LB got 0.57+",
      "votes": null
    },
    {
      "id": "1680565",
      "postDate": "02/07/2022 22:45:36",
      "content": "<p>Noob question - what does it mean to scale resolution beyond the 1280 width of the training images?  Is it just doing some kind of interpolation?    And if so why does it have such a big impact on performance?  </p>",
      "rawMarkdown": "Noob question - what does it mean to scale resolution beyond the 1280 width of the training images?  Is it just doing some kind of interpolation?    And if so why does it have such a big impact on performance?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1656714,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "01/19/2022 15:05:48",
      "content": "<p>It depends on scenarios.  if you read other hotness discussion, you would find that in crease size to  around 2x 2.5x , pretty good LB.    in other situations, maybe reduce image size.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1660004,
      "author_name": "geoffreygarrett",
      "author_url": "",
      "post_date": "01/22/2022 09:56:48",
      "content": "<p>HI coallar, I ended up using a resolution of 3072x3072, it was the largest I could go with my GPU (1080TI). I unfortunately don't have a 3090 <a href=\"https://www.kaggle.com/steamedsheep/yolov5-high-resolution-training\" target=\"_blank\">like sheep</a> 😂</p>",
      "votes": null,
      "replies": [
        {
          "id": 1660011,
          "author_name": "gjzhongdf163com",
          "author_url": "",
          "post_date": "01/22/2022 10:10:18",
          "content": "<p>my team partner uses 2080Ti, the resolution of 3600 was used, which LB got 0.57+</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1680565,
      "author_name": "max237",
      "author_url": "",
      "post_date": "02/07/2022 22:45:36",
      "content": "<p>Noob question - what does it mean to scale resolution beyond the 1280 width of the training images?  Is it just doing some kind of interpolation?    And if so why does it have such a big impact on performance?  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1656016": "Hi，every, resolution 1280 in this competition seems to get 0.4+ LB score, if double of it, 0.5+ will be got. So, what resolution did you use？",
    "1656714": "It depends on scenarios.  if you read other hotness discussion, you would find that in crease size to  around 2x 2.5x , pretty good LB.    in other situations, maybe reduce image size.",
    "1660004": "HI coallar, I ended up using a resolution of 3072x3072, it was the largest I could go with my GPU (1080TI). I unfortunately don't have a 3090 [like sheep](https://www.kaggle.com/steamedsheep/yolov5-high-resolution-training) 😂",
    "1660011": "my team partner uses 2080Ti, the resolution of 3600 was used, which LB got 0.57+",
    "1680565": "Noob question - what does it mean to scale resolution beyond the 1280 width of the training images?  Is it just doing some kind of interpolation?    And if so why does it have such a big impact on performance?"
  },
  "source": "meta"
}