{
  "id": 424956,
  "title": "What is the maximum score for a yolo model?LB：0.514",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/424956",
  "author_name": "bent1e",
  "post_date": "2023-07-16T13:40:42.162000",
  "votes": 8,
  "comment_count": 18,
  "views": 0,
  "content": "<p>yolov7（single modle）LB：0.514</p>",
  "messages": [
    {
      "id": 2346746,
      "postDate": "2023-07-16T13:40:42.163Z",
      "content": "<p>yolov7（single modle）LB：0.514</p>",
      "rawMarkdown": "yolov7（single modle）LB：0.514",
      "votes": 8
    },
    {
      "id": 2351352,
      "postDate": "2023-07-20T04:39:07.180Z",
      "content": "<p>May I know what types of data augmentation you did? We tried many combinations but the results were not ideal.<br>\nThank you very much!</p>",
      "rawMarkdown": "May I know what types of data augmentation you did? We tried many combinations but the results were not ideal.\nThank you very much!",
      "replies": [
        {
          "id": 2351582,
          "postDate": "2023-07-20T08:46:43.693Z",
          "content": "<p>flip，rot90，cutout，mixup</p>",
          "rawMarkdown": "flip，rot90，cutout，mixup",
          "votes": 1,
          "replies": [
            {
              "id": 2352451,
              "postDate": "2023-07-21T02:02:50.143Z",
              "content": "<p>Same yolov7x and strong augmentation strategies, but only lb0.467😭. Can you share your dilation kernel size?</p>",
              "rawMarkdown": "Same yolov7x and strong augmentation strategies, but only lb0.467😭. Can you share your dilation kernel size?"
            },
            {
              "id": 2352696,
              "postDate": "2023-07-21T07:48:52.077Z",
              "content": "<p>每個模型都不一樣，你得不停嘗試</p>",
              "rawMarkdown": "每個模型都不一樣，你得不停嘗試",
              "votes": 1
            }
          ]
        }
      ]
    },
    {
      "id": 2350880,
      "postDate": "2023-07-19T15:48:33.177Z",
      "content": "<p>Wow, cool. It is really interesting to know your scheme after the end of the competition.<br>\nI got 0.468 with MaskRCNN, but with Yolov8 I got only 0.327, but I tried to use different parameters.<br>\nAnd in my case Yolo has stopped learning after 12 epochs. I was stuck.</p>",
      "rawMarkdown": "Wow, cool. It is really interesting to know your scheme after the end of the competition.\nI got 0.468 with MaskRCNN, but with Yolov8 I got only 0.327, but I tried to use different parameters.\nAnd in my case Yolo has stopped learning after 12 epochs. I was stuck.",
      "replies": [
        {
          "id": 2351260,
          "postDate": "2023-07-20T02:05:14.397Z",
          "content": "<p>我和你相反，我使用MaskRCNN只能得到0.435的分數</p>",
          "rawMarkdown": "我和你相反，我使用MaskRCNN只能得到0.435的分數"
        }
      ]
    },
    {
      "id": 2350485,
      "postDate": "2023-07-19T09:03:56.827Z",
      "content": "<p>For me, only 0.46.  Did you use pseudo labels or modified labels on dataset2?</p>",
      "rawMarkdown": "For me, only 0.46.  Did you use pseudo labels or modified labels on dataset2?",
      "replies": [
        {
          "id": 2350536,
          "postDate": "2023-07-19T09:32:57.867Z",
          "content": "<p>沒有，我只在dataset3上使用了偽標簽，但并不起作用</p>",
          "rawMarkdown": "沒有，我只在dataset3上使用了偽標簽，但并不起作用",
          "votes": 1,
          "replies": [
            {
              "id": 2350635,
              "postDate": "2023-07-19T11:17:26.273Z",
              "content": "<p>请问你是单纯使用yolov7就能达到0.5的效果吗？还是有额外对数据集进行处理后才有达到这种水平？</p>",
              "rawMarkdown": "请问你是单纯使用yolov7就能达到0.5的效果吗？还是有额外对数据集进行处理后才有达到这种水平？"
            },
            {
              "id": 2350904,
              "postDate": "2023-07-19T16:08:01.530Z",
              "content": "<p>感谢回答！这将对我很有帮助！</p>",
              "rawMarkdown": "感谢回答！这将对我很有帮助！"
            },
            {
              "id": 2351254,
              "postDate": "2023-07-20T02:03:01.787Z",
              "content": "<p>如果你將數據增强理解成額外處理，那就是有</p>",
              "rawMarkdown": "如果你將數據增强理解成額外處理，那就是有"
            }
          ]
        },
        {
          "id": 2350634,
          "postDate": "2023-07-19T11:16:01.530Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 2347465,
      "postDate": "2023-07-17T02:05:53.927Z",
      "content": "<p>for me, it's 0.416, which are you using, s/m/l?</p>",
      "rawMarkdown": "for me, it's 0.416, which are you using, s/m/l?",
      "replies": [
        {
          "id": 2349449,
          "postDate": "2023-07-18T12:39:29.370Z",
          "content": "<p>which are you using? I use s and my result is only 0.388.  Whether to modify network parameters？</p>",
          "rawMarkdown": "which are you using? I use s and my result is only 0.388.  Whether to modify network parameters？",
          "replies": [
            {
              "id": 2351290,
              "postDate": "2023-07-20T03:05:03.537Z",
              "content": "<p>yolov8s, default params</p>",
              "rawMarkdown": "yolov8s, default params"
            }
          ]
        },
        {
          "id": 2350321,
          "postDate": "2023-07-19T07:01:42.937Z",
          "content": "<p>i use yolov7x</p>",
          "rawMarkdown": "i use yolov7x",
          "votes": 1
        }
      ]
    },
    {
      "id": 2346927,
      "postDate": "2023-07-16T16:22:39.237Z",
      "content": "<p>Is it for one fold?</p>",
      "rawMarkdown": "Is it for one fold?",
      "replies": [
        {
          "id": 2350322,
          "postDate": "2023-07-19T07:02:31.010Z",
          "content": "<p>yes,model ensemble is not work for me</p>",
          "rawMarkdown": "yes,model ensemble is not work for me",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2351352,
      "author_name": "zwtunsw",
      "author_url": "",
      "post_date": "2023-07-20T04:39:07.180000",
      "content": "<p>May I know what types of data augmentation you did? We tried many combinations but the results were not ideal.<br>\nThank you very much!</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2351582,
          "author_name": "bent1e",
          "author_url": "",
          "post_date": "2023-07-20T08:46:43.693000",
          "content": "<p>flip，rot90，cutout，mixup</p>",
          "votes": 1,
          "replies": [
            {
              "id": 2352451,
              "author_name": "RickyLu",
              "author_url": "",
              "post_date": "2023-07-21T02:02:50.143000",
              "content": "<p>Same yolov7x and strong augmentation strategies, but only lb0.467😭. Can you share your dilation kernel size?</p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2352696,
              "author_name": "bent1e",
              "author_url": "",
              "post_date": "2023-07-21T07:48:52.077000",
              "content": "<p>每個模型都不一樣，你得不停嘗試</p>",
              "votes": 1,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2350880,
      "author_name": "Jura Moshkov",
      "author_url": "",
      "post_date": "2023-07-19T15:48:33.177000",
      "content": "<p>Wow, cool. It is really interesting to know your scheme after the end of the competition.<br>\nI got 0.468 with MaskRCNN, but with Yolov8 I got only 0.327, but I tried to use different parameters.<br>\nAnd in my case Yolo has stopped learning after 12 epochs. I was stuck.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2351260,
          "author_name": "bent1e",
          "author_url": "",
          "post_date": "2023-07-20T02:05:14.397000",
          "content": "<p>我和你相反，我使用MaskRCNN只能得到0.435的分數</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2350485,
      "author_name": "FreddyHu",
      "author_url": "",
      "post_date": "2023-07-19T09:03:56.827000",
      "content": "<p>For me, only 0.46.  Did you use pseudo labels or modified labels on dataset2?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2350536,
          "author_name": "bent1e",
          "author_url": "",
          "post_date": "2023-07-19T09:32:57.867000",
          "content": "<p>沒有，我只在dataset3上使用了偽標簽，但并不起作用</p>",
          "votes": 1,
          "replies": [
            {
              "id": 2350635,
              "author_name": "xiaoxu Xu",
              "author_url": "",
              "post_date": "2023-07-19T11:17:26.273000",
              "content": "<p>请问你是单纯使用yolov7就能达到0.5的效果吗？还是有额外对数据集进行处理后才有达到这种水平？</p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2350904,
              "author_name": "FreddyHu",
              "author_url": "",
              "post_date": "2023-07-19T16:08:01.530000",
              "content": "<p>感谢回答！这将对我很有帮助！</p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2351254,
              "author_name": "bent1e",
              "author_url": "",
              "post_date": "2023-07-20T02:03:01.787000",
              "content": "<p>如果你將數據增强理解成額外處理，那就是有</p>",
              "votes": 0,
              "replies": []
            }
          ]
        },
        {
          "id": 2350634,
          "author_name": "",
          "author_url": "",
          "post_date": "2023-07-19T11:16:01.530000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2347465,
      "author_name": "Yi Wu",
      "author_url": "",
      "post_date": "2023-07-17T02:05:53.927000",
      "content": "<p>for me, it's 0.416, which are you using, s/m/l?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2349449,
          "author_name": "xiaoxu Xu",
          "author_url": "",
          "post_date": "2023-07-18T12:39:29.370000",
          "content": "<p>which are you using? I use s and my result is only 0.388.  Whether to modify network parameters？</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2351290,
              "author_name": "Yi Wu",
              "author_url": "",
              "post_date": "2023-07-20T03:05:03.537000",
              "content": "<p>yolov8s, default params</p>",
              "votes": 0,
              "replies": []
            }
          ]
        },
        {
          "id": 2350321,
          "author_name": "bent1e",
          "author_url": "",
          "post_date": "2023-07-19T07:01:42.937000",
          "content": "<p>i use yolov7x</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 2346927,
      "author_name": "Araik Tamazian",
      "author_url": "",
      "post_date": "2023-07-16T16:22:39.237000",
      "content": "<p>Is it for one fold?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2350322,
          "author_name": "bent1e",
          "author_url": "",
          "post_date": "2023-07-19T07:02:31.010000",
          "content": "<p>yes,model ensemble is not work for me</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2346746": "yolov7（single modle）LB：0.514",
    "2351352": "May I know what types of data augmentation you did? We tried many combinations but the results were not ideal.\nThank you very much!",
    "2350880": "Wow, cool. It is really interesting to know your scheme after the end of the competition.\nI got 0.468 with MaskRCNN, but with Yolov8 I got only 0.327, but I tried to use different parameters.\nAnd in my case Yolo has stopped learning after 12 epochs. I was stuck.",
    "2350485": "For me, only 0.46.  Did you use pseudo labels or modified labels on dataset2?",
    "2347465": "for me, it's 0.416, which are you using, s/m/l?",
    "2346927": "Is it for one fold?"
  }
}