{
  "id": 263366,
  "title": "There are attempts to fuse detection models",
  "url": "/competitions/siim-covid19-detection/discussion/263366",
  "author_name": "",
  "post_date": "2021-08-09T06:41:43.817317800Z",
  "votes": 1,
  "comment_count": 12,
  "views": 0,
  "content": "<p>I tried to use WBF to fuse the frames of two models with the same image, but failed every time, is there any good method</p>",
  "messages": [
    {
      "id": "1460950",
      "postDate": "08/09/2021 06:41:43",
      "content": "<p>I tried to use WBF to fuse the frames of two models with the same image, but failed every time, is there any good method</p>",
      "rawMarkdown": "I tried to use WBF to fuse the frames of two models with the same image, but failed every time, is there any good method",
      "votes": null
    },
    {
      "id": "1461074",
      "postDate": "08/09/2021 07:47:44",
      "content": "<p>everything is written there:<br>\nweights = None #[2,1]<br>\niou_thr = ? #0.5<br>\nskip_box_thr = ? #0.0001<br>\nboxes, scores, labels = weighted_boxes_fusion([boxes_list, boxes_list1],[scores_list, scores_list1],[labels_list, labels_list1]) weights=weights, iou_thr=iou_thr, skip_box_thr=skip_box_thr)</p>",
      "rawMarkdown": "everything is written there:\nweights = None #[2,1]\niou_thr = ? #0.5\nskip_box_thr = ? #0.0001\nboxes, scores, labels = weighted_boxes_fusion([boxes_list, boxes_list1],[scores_list, scores_list1],[labels_list, labels_list1]) weights=weights, iou_thr=iou_thr, skip_box_thr=skip_box_thr)",
      "votes": null
    },
    {
      "id": "1461088",
      "postDate": "08/09/2021 07:57:18",
      "content": "<p>thank you<br>\nI have tried this method to fuse the boxes on each image, but there is no result after fusion. I don't know why.<br>\nIs there any way I can improve by 0.002, I want a bronze medal, but I can't think of anything right now</p>",
      "rawMarkdown": "thank you\nI have tried this method to fuse the boxes on each image, but there is no result after fusion. I don't know why.\nIs there any way I can improve by 0.002, I want a bronze medal, but I can't think of anything right now",
      "votes": null
    },
    {
      "id": "1461324",
      "postDate": "08/09/2021 10:49:08",
      "content": "<p>I do not know why you have no result. Maybe something with the models or with the code. The improvement method is possible and there, but it would be, at least, not correct to voice it. all of us can only hope for a reassessment of the result tomorrow. Everything can change.</p>",
      "rawMarkdown": "I do not know why you have no result. Maybe something with the models or with the code. The improvement method is possible and there, but it would be, at least, not correct to voice it. all of us can only hope for a reassessment of the result tomorrow. Everything can change.",
      "votes": null
    },
    {
      "id": "1462295",
      "postDate": "08/09/2021 19:13:52",
      "content": "<p>Did you normalize to box to range of 0 and 1?</p>",
      "rawMarkdown": "Did you normalize to box to range of 0 and 1?",
      "votes": null
    },
    {
      "id": "1462602",
      "postDate": "08/09/2021 23:34:06",
      "content": "<p>It did yesterday because there was no normalization，but It didn't work very well</p>",
      "rawMarkdown": "It did yesterday because there was no normalization，but It didn't work very well",
      "votes": null
    },
    {
      "id": "1463461",
      "postDate": "08/10/2021 07:48:58",
      "content": "<p>You see - we are next to you. I chose the wrong applications.</p>",
      "rawMarkdown": "You see - we are next to you. I chose the wrong applications.",
      "votes": null
    },
    {
      "id": "1463618",
      "postDate": "08/10/2021 08:57:51",
      "content": "<p>The result of one of our programs reached 0.601, but I didn't choose it because the LB score was low</p>",
      "rawMarkdown": "The result of one of our programs reached 0.601, but I didn't choose it because the LB score was low",
      "votes": null
    },
    {
      "id": "1463692",
      "postDate": "08/10/2021 09:33:59",
      "content": "<p>I have only 300 applications. maximum 604 - pieces 10, 602, 601 I do not count at all - it turned out funny. Good luck. <a href=\"https://prnt.sc/1m6uh7i\" target=\"_blank\">https://prnt.sc/1m6uh7i</a></p>",
      "rawMarkdown": "I have only 300 applications. maximum 604 - pieces 10, 602, 601 I do not count at all - it turned out funny. Good luck. https://prnt.sc/1m6uh7i",
      "votes": null
    },
    {
      "id": "1463717",
      "postDate": "08/10/2021 09:47:18",
      "content": "<p>What a pity! How did you get 0.629 LB</p>",
      "rawMarkdown": "What a pity! How did you get 0.629 LB",
      "votes": null
    },
    {
      "id": "1463736",
      "postDate": "08/10/2021 09:58:21",
      "content": "<p>classifier ensemble + detector ensemble yolov5 and mmdetection (2 models, 4 and 3) - without any special bells and whistles and third-party data, programs and other techniques. what did you use</p>",
      "rawMarkdown": "classifier ensemble + detector ensemble yolov5 and mmdetection (2 models, 4 and 3) - without any special bells and whistles and third-party data, programs and other techniques. what did you use",
      "votes": null
    },
    {
      "id": "1463928",
      "postDate": "08/10/2021 11:41:44",
      "content": "<p>Classified V2+B7, detected Yolov5 and MMdetection, but I detected very poor fusion score using WBF</p>",
      "rawMarkdown": "Classified V2+B7, detected Yolov5 and MMdetection, but I detected very poor fusion score using WBF",
      "votes": null
    },
    {
      "id": "1464095",
      "postDate": "08/10/2021 12:54:39",
      "content": "<p>it turns out that me too</p>",
      "rawMarkdown": "it turns out that me too",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1461074,
      "author_name": "zaakciiru",
      "author_url": "",
      "post_date": "08/09/2021 07:47:44",
      "content": "<p>everything is written there:<br>\nweights = None #[2,1]<br>\niou_thr = ? #0.5<br>\nskip_box_thr = ? #0.0001<br>\nboxes, scores, labels = weighted_boxes_fusion([boxes_list, boxes_list1],[scores_list, scores_list1],[labels_list, labels_list1]) weights=weights, iou_thr=iou_thr, skip_box_thr=skip_box_thr)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1461088,
          "author_name": "dlidli",
          "author_url": "",
          "post_date": "08/09/2021 07:57:18",
          "content": "<p>thank you<br>\nI have tried this method to fuse the boxes on each image, but there is no result after fusion. I don't know why.<br>\nIs there any way I can improve by 0.002, I want a bronze medal, but I can't think of anything right now</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1461324,
          "author_name": "zaakciiru",
          "author_url": "",
          "post_date": "08/09/2021 10:49:08",
          "content": "<p>I do not know why you have no result. Maybe something with the models or with the code. The improvement method is possible and there, but it would be, at least, not correct to voice it. all of us can only hope for a reassessment of the result tomorrow. Everything can change.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1462295,
          "author_name": "sunpnwt12",
          "author_url": "",
          "post_date": "08/09/2021 19:13:52",
          "content": "<p>Did you normalize to box to range of 0 and 1?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1462602,
          "author_name": "dlidli",
          "author_url": "",
          "post_date": "08/09/2021 23:34:06",
          "content": "<p>It did yesterday because there was no normalization，but It didn't work very well</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1463461,
      "author_name": "zaakciiru",
      "author_url": "",
      "post_date": "08/10/2021 07:48:58",
      "content": "<p>You see - we are next to you. I chose the wrong applications.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1463618,
          "author_name": "dlidli",
          "author_url": "",
          "post_date": "08/10/2021 08:57:51",
          "content": "<p>The result of one of our programs reached 0.601, but I didn't choose it because the LB score was low</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1463692,
          "author_name": "zaakciiru",
          "author_url": "",
          "post_date": "08/10/2021 09:33:59",
          "content": "<p>I have only 300 applications. maximum 604 - pieces 10, 602, 601 I do not count at all - it turned out funny. Good luck. <a href=\"https://prnt.sc/1m6uh7i\" target=\"_blank\">https://prnt.sc/1m6uh7i</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1463717,
          "author_name": "dlidli",
          "author_url": "",
          "post_date": "08/10/2021 09:47:18",
          "content": "<p>What a pity! How did you get 0.629 LB</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1463736,
          "author_name": "zaakciiru",
          "author_url": "",
          "post_date": "08/10/2021 09:58:21",
          "content": "<p>classifier ensemble + detector ensemble yolov5 and mmdetection (2 models, 4 and 3) - without any special bells and whistles and third-party data, programs and other techniques. what did you use</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1463928,
          "author_name": "dlidli",
          "author_url": "",
          "post_date": "08/10/2021 11:41:44",
          "content": "<p>Classified V2+B7, detected Yolov5 and MMdetection, but I detected very poor fusion score using WBF</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1464095,
          "author_name": "zaakciiru",
          "author_url": "",
          "post_date": "08/10/2021 12:54:39",
          "content": "<p>it turns out that me too</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1460950": "I tried to use WBF to fuse the frames of two models with the same image, but failed every time, is there any good method",
    "1461074": "everything is written there:\nweights = None #[2,1]\niou_thr = ? #0.5\nskip_box_thr = ? #0.0001\nboxes, scores, labels = weighted_boxes_fusion([boxes_list, boxes_list1],[scores_list, scores_list1],[labels_list, labels_list1]) weights=weights, iou_thr=iou_thr, skip_box_thr=skip_box_thr)",
    "1461088": "thank you\nI have tried this method to fuse the boxes on each image, but there is no result after fusion. I don't know why.\nIs there any way I can improve by 0.002, I want a bronze medal, but I can't think of anything right now",
    "1461324": "I do not know why you have no result. Maybe something with the models or with the code. The improvement method is possible and there, but it would be, at least, not correct to voice it. all of us can only hope for a reassessment of the result tomorrow. Everything can change.",
    "1462295": "Did you normalize to box to range of 0 and 1?",
    "1462602": "It did yesterday because there was no normalization，but It didn't work very well",
    "1463461": "You see - we are next to you. I chose the wrong applications.",
    "1463618": "The result of one of our programs reached 0.601, but I didn't choose it because the LB score was low",
    "1463692": "I have only 300 applications. maximum 604 - pieces 10, 602, 601 I do not count at all - it turned out funny. Good luck. https://prnt.sc/1m6uh7i",
    "1463717": "What a pity! How did you get 0.629 LB",
    "1463736": "classifier ensemble + detector ensemble yolov5 and mmdetection (2 models, 4 and 3) - without any special bells and whistles and third-party data, programs and other techniques. what did you use",
    "1463928": "Classified V2+B7, detected Yolov5 and MMdetection, but I detected very poor fusion score using WBF",
    "1464095": "it turns out that me too"
  },
  "source": "meta"
}