{
  "id": 420817,
  "title": "[0.412] How can I improve my Score? ",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/420817",
  "author_name": "",
  "post_date": "2023-07-02T17:35:25.882118600Z",
  "votes": 7,
  "comment_count": 5,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9770977%2Fc2d2e611e6f6b50f4b8b4b271de74690%2FScreenshot%202023-07-03%20at%2000.25.08.png?generation=1688318748130585&amp;alt=media\" alt=\"\"></p>\n<p>I joined the contest for 3 days.<br>\nI started by looking at some yolov7 and yolov8 notebooks</p>\n<ul>\n<li><p>yolov7: I started with model yolov7, the results are quite good, with lb .<br>\nthen I adjusted a bit in the model and changed to another data but the result didn't improve much</p></li>\n<li><p>yolov8: same result as yolov7. and I stopped with this pattern</p></li>\n<li><p>Mmdet 3.0 : lb 0.412 the first time I tested it. Then I tried to train model at size 784 and test with size 512 and it failed. I continue to experiment with MaskRCNN101 but the result is only 0.396</p></li>\n</ul>\n<p>is there any way i can continue to improve my results with yolov7 and Mmdet 3.0 ?</p>",
  "messages": [
    {
      "id": "2327166",
      "postDate": "07/02/2023 17:35:25",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9770977%2Fc2d2e611e6f6b50f4b8b4b271de74690%2FScreenshot%202023-07-03%20at%2000.25.08.png?generation=1688318748130585&amp;alt=media\" alt=\"\"></p>\n<p>I joined the contest for 3 days.<br>\nI started by looking at some yolov7 and yolov8 notebooks</p>\n<ul>\n<li><p>yolov7: I started with model yolov7, the results are quite good, with lb .<br>\nthen I adjusted a bit in the model and changed to another data but the result didn't improve much</p></li>\n<li><p>yolov8: same result as yolov7. and I stopped with this pattern</p></li>\n<li><p>Mmdet 3.0 : lb 0.412 the first time I tested it. Then I tried to train model at size 784 and test with size 512 and it failed. I continue to experiment with MaskRCNN101 but the result is only 0.396</p></li>\n</ul>\n<p>is there any way i can continue to improve my results with yolov7 and Mmdet 3.0 ?</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9770977%2Fc2d2e611e6f6b50f4b8b4b271de74690%2FScreenshot%202023-07-03%20at%2000.25.08.png?generation=1688318748130585&alt=media)\n\nI joined the contest for 3 days.\nI started by looking at some yolov7 and yolov8 notebooks\n\n- yolov7: I started with model yolov7, the results are quite good, with lb .\nthen I adjusted a bit in the model and changed to another data but the result didn't improve much\n- yolov8: same result as yolov7. and I stopped with this pattern\n\n- Mmdet 3.0 : lb 0.412 the first time I tested it. Then I tried to train model at size 784 and test with size 512 and it failed. I continue to experiment with MaskRCNN101 but the result is only 0.396\n\nis there any way i can continue to improve my results with yolov7 and Mmdet 3.0 ?",
      "votes": null
    },
    {
      "id": "2332024",
      "postDate": "07/05/2023 23:26:20",
      "content": "<p>Did you use some post processing methods? <br>\nDilation, tta?</p>",
      "rawMarkdown": "Did you use some post processing methods? \nDilation, tta?",
      "votes": null
    },
    {
      "id": "2347797",
      "postDate": "07/17/2023 07:50:00",
      "content": "<p>Can I use dilation or tta on my train-set before training?<br>\nActually I have not acknowledge how the  post processing methods works.<br>\nIt's my honor If you can give me some advice .</p>",
      "rawMarkdown": "Can I use dilation or tta on my train-set before training?\nActually I have not acknowledge how the  post processing methods works.\nIt's my honor If you can give me some advice .",
      "votes": null
    },
    {
      "id": "2348274",
      "postDate": "07/17/2023 14:05:16",
      "content": "<p>For me, using dilation in the training set doesn't help</p>",
      "rawMarkdown": "For me, using dilation in the training set doesn't help",
      "votes": null
    },
    {
      "id": "2348287",
      "postDate": "07/17/2023 14:12:47",
      "content": "<p>I think so after having trained.</p>",
      "rawMarkdown": "I think so after having trained.",
      "votes": null
    },
    {
      "id": "2351591",
      "postDate": "07/20/2023 08:53:24",
      "content": "<p>just dilate dataset2 might help a bit</p>",
      "rawMarkdown": "just dilate dataset2 might help a bit",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2332024,
      "author_name": "climbest",
      "author_url": "",
      "post_date": "07/05/2023 23:26:20",
      "content": "<p>Did you use some post processing methods? <br>\nDilation, tta?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2347797,
          "author_name": "magicowl",
          "author_url": "",
          "post_date": "07/17/2023 07:50:00",
          "content": "<p>Can I use dilation or tta on my train-set before training?<br>\nActually I have not acknowledge how the  post processing methods works.<br>\nIt's my honor If you can give me some advice .</p>",
          "votes": null,
          "replies": [
            {
              "id": 2348274,
              "author_name": "dav1d7",
              "author_url": "",
              "post_date": "07/17/2023 14:05:16",
              "content": "<p>For me, using dilation in the training set doesn't help</p>",
              "votes": null,
              "replies": [
                {
                  "id": 2348287,
                  "author_name": "magicowl",
                  "author_url": "",
                  "post_date": "07/17/2023 14:12:47",
                  "content": "<p>I think so after having trained.</p>",
                  "votes": null,
                  "replies": [
                    {
                      "id": 2351591,
                      "author_name": "bent1e",
                      "author_url": "",
                      "post_date": "07/20/2023 08:53:24",
                      "content": "<p>just dilate dataset2 might help a bit</p>",
                      "votes": null,
                      "replies": []
                    }
                  ]
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2327166": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9770977%2Fc2d2e611e6f6b50f4b8b4b271de74690%2FScreenshot%202023-07-03%20at%2000.25.08.png?generation=1688318748130585&alt=media)\n\nI joined the contest for 3 days.\nI started by looking at some yolov7 and yolov8 notebooks\n\n- yolov7: I started with model yolov7, the results are quite good, with lb .\nthen I adjusted a bit in the model and changed to another data but the result didn't improve much\n- yolov8: same result as yolov7. and I stopped with this pattern\n\n- Mmdet 3.0 : lb 0.412 the first time I tested it. Then I tried to train model at size 784 and test with size 512 and it failed. I continue to experiment with MaskRCNN101 but the result is only 0.396\n\nis there any way i can continue to improve my results with yolov7 and Mmdet 3.0 ?",
    "2332024": "Did you use some post processing methods? \nDilation, tta?",
    "2347797": "Can I use dilation or tta on my train-set before training?\nActually I have not acknowledge how the  post processing methods works.\nIt's my honor If you can give me some advice .",
    "2348274": "For me, using dilation in the training set doesn't help",
    "2348287": "I think so after having trained.",
    "2351591": "just dilate dataset2 might help a bit"
  },
  "source": "meta"
}