{
  "id": 263701,
  "title": "11th place solution",
  "url": "/competitions/siim-covid19-detection/writeups/11th-place-solution",
  "author_name": "",
  "post_date": "2021-08-12T00:49:11.983Z",
  "votes": 26,
  "comment_count": 11,
  "views": 0,
  "content": "<p>First of all, I would like to thank you for organizing this very interesting competition and congratulations to all the winners of this competition. Also thanks to everyone in our team <a href=\"https://www.kaggle.com/quochungto\" target=\"_blank\">@quochungto</a> <a href=\"https://www.kaggle.com/morizin\" target=\"_blank\">@morizin</a> <a href=\"https://www.kaggle.com/joven1997\" target=\"_blank\">@joven1997</a> <a href=\"https://www.kaggle.com/socom20\" target=\"_blank\">@socom20</a> for allowing me to win the first gold medal in this competition. </p>\n<p><strong><em>Study Level</em></strong></p>\n<ul>\n<li>Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) with siim covid dataset .</li>\n<li>Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) + pseudo label with siim covid dataset + test dataset + ricord dataset + bimcv dataset.（When dealing with bimcv external data, I would like to thank <a href=\"https://www.kaggle.com/quochungto\" target=\"_blank\">@quochungto</a> and <a href=\"https://www.kaggle.com/joven1997\" target=\"_blank\">@joven1997</a> manually deleting the noise data.）</li>\n<li>Augmentation: Resize, HorizontalFlip, RandomBrightness, RandomContrast, OpticalDistortion, GridDistortion, HueSaturationValue, ShiftScaleRotate, CutoutV2.</li>\n<li>Loss:  <br>\ncls loss = BiTemperedLogisticLoss  <a href=\"https://github.com/mlpanda/bi-tempered-loss-pytorch/blob/master/bi_tempered_loss.py\" target=\"_blank\">link</a>.<br>\naux loss = 0.9<em>bce + 0.1</em>lovasz.</li>\n<li>Optimizer: Adam with init_learning_rate 0.0001.</li>\n<li>Scheduler: GradualWarmupSchedulerV2 <a href=\"https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver\" target=\"_blank\">link</a>.</li>\n<li>Test time augmentation: src+ horizontal flip.</li>\n<li>Final submission includes 4 models with diversity of encoders, decoders, input size:<br>\nEfficientnet_b5 512, Efficientnet_b7 512, Efficientnetv2_m 512, Swim-Transformer 384.</li>\n<li>Ensemble method: more details <a href=\"https://www.kaggle.com/cdeotte/forward-selection-oof-ensemble-0-942-private\" target=\"_blank\">link</a> (Thanks for <a href=\"https://www.kaggle.com/morizin\" target=\"_blank\">@morizin</a> lot of ideas)</li>\n</ul>\n<p>study level part source code used in this competition  has been uploaded to the following github:<br>\n<a href=\"https://github.com/ChenYingpeng/pl-siim-covid19-detection\" target=\"_blank\">https://github.com/ChenYingpeng/pl-siim-covid19-detection</a></p>\n<ul>\n<li><p>What didn't work<br>\n Segmentation channel images as RANZCR competition<br>\n using more TTA(such as scale=1.25, vertical flip etc)<br>\n Training 2 classes (none and opacity) in study level.<br>\n Training  2 classes (none and opacity) in image level. （Use the entire image like none as its box）</p>\n<table>\n<thead>\n<tr>\n<th>arch</th>\n<th>cv (5-folds mAP*(4/6))</th>\n<th>lb</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>effnet-b5</td>\n<td>0.387736</td>\n<td>0.535</td>\n</tr>\n<tr>\n<td>effnet-b5-pseudo</td>\n<td>0.3925984</td>\n<td>---</td>\n</tr>\n<tr>\n<td>effnet-b7</td>\n<td>0.3821176</td>\n<td>0.530</td>\n</tr>\n<tr>\n<td>effnetv2-m</td>\n<td>0.3869362</td>\n<td>0.532</td>\n</tr>\n<tr>\n<td>swim-transformer</td>\n<td>0.3815662</td>\n<td>---</td>\n</tr>\n</tbody>\n</table></li>\n</ul>\n<p><strong><em>Image Level</em></strong></p>\n<ul>\n<li>YOLOv5 <br>\nyolotrs-384, yolotrl-384,yolov5x-384, yolov5x-512, yolov5x-640.<br>\nI just do yolotrs-384 and yolotrs-384+pseudo with test dataset in image level part.<br>\nyolov5s + transformer  <a href=\"https://www.gitmemory.com/issue/ultralytics/yolov5/2329/808852867\" target=\"_blank\">link</a><br>\ndetection mAP@0.5 opacity class.</li>\n</ul>\n<table>\n<thead>\n<tr>\n<th></th>\n<th>yolotrs-384</th>\n<th>yolotrs-384-pseudo</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>mAP0.5 TTA</td>\n<td>0.567</td>\n<td>0.576</td>\n</tr>\n<tr>\n<td>LB</td>\n<td>0.092</td>\n<td>0.095</td>\n</tr>\n</tbody>\n</table>\n<p>For more details, please see <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/264287\" target=\"_blank\">https://www.kaggle.com/c/siim-covid19-detection/discussion/264287</a></p>\n<p><strong><em>Final Submission</em></strong><br>\nFinal Submission Notebook: <a href=\"https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292\" target=\"_blank\">https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292</a></p>",
  "messages": [
    {
      "id": "1462941",
      "postDate": "08/10/2021 03:42:30",
      "content": "<p>First of all, I would like to thank you for organizing this very interesting competition and congratulations to all the winners of this competition. Also thanks to everyone in our team <a href=\"https://www.kaggle.com/quochungto\" target=\"_blank\">@quochungto</a> <a href=\"https://www.kaggle.com/morizin\" target=\"_blank\">@morizin</a> <a href=\"https://www.kaggle.com/joven1997\" target=\"_blank\">@joven1997</a> <a href=\"https://www.kaggle.com/socom20\" target=\"_blank\">@socom20</a> for allowing me to win the first gold medal in this competition. </p>\n<p><strong><em>Study Level</em></strong></p>\n<ul>\n<li>Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) with siim covid dataset .</li>\n<li>Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) + pseudo label with siim covid dataset + test dataset + ricord dataset + bimcv dataset.（When dealing with bimcv external data, I would like to thank <a href=\"https://www.kaggle.com/quochungto\" target=\"_blank\">@quochungto</a> and <a href=\"https://www.kaggle.com/joven1997\" target=\"_blank\">@joven1997</a> manually deleting the noise data.）</li>\n<li>Augmentation: Resize, HorizontalFlip, RandomBrightness, RandomContrast, OpticalDistortion, GridDistortion, HueSaturationValue, ShiftScaleRotate, CutoutV2.</li>\n<li>Loss:  <br>\ncls loss = BiTemperedLogisticLoss  <a href=\"https://github.com/mlpanda/bi-tempered-loss-pytorch/blob/master/bi_tempered_loss.py\" target=\"_blank\">link</a>.<br>\naux loss = 0.9<em>bce + 0.1</em>lovasz.</li>\n<li>Optimizer: Adam with init_learning_rate 0.0001.</li>\n<li>Scheduler: GradualWarmupSchedulerV2 <a href=\"https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver\" target=\"_blank\">link</a>.</li>\n<li>Test time augmentation: src+ horizontal flip.</li>\n<li>Final submission includes 4 models with diversity of encoders, decoders, input size:<br>\nEfficientnet_b5 512, Efficientnet_b7 512, Efficientnetv2_m 512, Swim-Transformer 384.</li>\n<li>Ensemble method: more details <a href=\"https://www.kaggle.com/cdeotte/forward-selection-oof-ensemble-0-942-private\" target=\"_blank\">link</a> (Thanks for <a href=\"https://www.kaggle.com/morizin\" target=\"_blank\">@morizin</a> lot of ideas)</li>\n</ul>\n<p>study level part source code used in this competition  has been uploaded to the following github:<br>\n<a href=\"https://github.com/ChenYingpeng/pl-siim-covid19-detection\" target=\"_blank\">https://github.com/ChenYingpeng/pl-siim-covid19-detection</a></p>\n<ul>\n<li><p>What didn't work<br>\n Segmentation channel images as RANZCR competition<br>\n using more TTA(such as scale=1.25, vertical flip etc)<br>\n Training 2 classes (none and opacity) in study level.<br>\n Training  2 classes (none and opacity) in image level. （Use the entire image like none as its box）</p>\n<table>\n<thead>\n<tr>\n<th>arch</th>\n<th>cv (5-folds mAP*(4/6))</th>\n<th>lb</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>effnet-b5</td>\n<td>0.387736</td>\n<td>0.535</td>\n</tr>\n<tr>\n<td>effnet-b5-pseudo</td>\n<td>0.3925984</td>\n<td>---</td>\n</tr>\n<tr>\n<td>effnet-b7</td>\n<td>0.3821176</td>\n<td>0.530</td>\n</tr>\n<tr>\n<td>effnetv2-m</td>\n<td>0.3869362</td>\n<td>0.532</td>\n</tr>\n<tr>\n<td>swim-transformer</td>\n<td>0.3815662</td>\n<td>---</td>\n</tr>\n</tbody>\n</table></li>\n</ul>\n<p><strong><em>Image Level</em></strong></p>\n<ul>\n<li>YOLOv5 <br>\nyolotrs-384, yolotrl-384,yolov5x-384, yolov5x-512, yolov5x-640.<br>\nI just do yolotrs-384 and yolotrs-384+pseudo with test dataset in image level part.<br>\nyolov5s + transformer  <a href=\"https://www.gitmemory.com/issue/ultralytics/yolov5/2329/808852867\" target=\"_blank\">link</a><br>\ndetection mAP@0.5 opacity class.</li>\n</ul>\n<table>\n<thead>\n<tr>\n<th></th>\n<th>yolotrs-384</th>\n<th>yolotrs-384-pseudo</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>mAP0.5 TTA</td>\n<td>0.567</td>\n<td>0.576</td>\n</tr>\n<tr>\n<td>LB</td>\n<td>0.092</td>\n<td>0.095</td>\n</tr>\n</tbody>\n</table>\n<p>For more details, please see <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/264287\" target=\"_blank\">https://www.kaggle.com/c/siim-covid19-detection/discussion/264287</a></p>\n<p><strong><em>Final Submission</em></strong><br>\nFinal Submission Notebook: <a href=\"https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292\" target=\"_blank\">https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292</a></p>",
      "rawMarkdown": "First of all, I would like to thank you for organizing this very interesting competition and congratulations to all the winners of this competition. Also thanks to everyone in our team @quochungto @morizin @joven1997 @socom20 for allowing me to win the first gold medal in this competition. \n\n***Study Level***\n\n- Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) with siim covid dataset .\n- Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) + pseudo label with siim covid dataset + test dataset + ricord dataset + bimcv dataset.（When dealing with bimcv external data, I would like to thank @quochungto and @joven1997 manually deleting the noise data.）\n- Augmentation: Resize, HorizontalFlip, RandomBrightness, RandomContrast, OpticalDistortion, GridDistortion, HueSaturationValue, ShiftScaleRotate, CutoutV2.\n- Loss:  \ncls loss = BiTemperedLogisticLoss  [link](https://github.com/mlpanda/bi-tempered-loss-pytorch/blob/master/bi_tempered_loss.py).\naux loss = 0.9*bce + 0.1*lovasz.\n- Optimizer: Adam with init_learning_rate 0.0001.\n- Scheduler: GradualWarmupSchedulerV2 [link](https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver).\n- Test time augmentation: src+ horizontal flip.\n- Final submission includes 4 models with diversity of encoders, decoders, input size:\nEfficientnet_b5 512, Efficientnet_b7 512, Efficientnetv2_m 512, Swim-Transformer 384.\n- Ensemble method: more details [link](https://www.kaggle.com/cdeotte/forward-selection-oof-ensemble-0-942-private) (Thanks for @morizin lot of ideas)\n\nstudy level part source code used in this competition  has been uploaded to the following github:\nhttps://github.com/ChenYingpeng/pl-siim-covid19-detection\n\n- What didn't work\n     Segmentation channel images as RANZCR competition\n     using more TTA(such as scale=1.25, vertical flip etc)\n     Training 2 classes (none and opacity) in study level.\n     Training  2 classes (none and opacity) in image level. （Use the entire image like none as its box）\n\n arch| cv (5-folds mAP*(4/6))| lb |\n| --- | --- | ---|\n| effnet-b5 |0.387736  | 0.535|\n| effnet-b5-pseudo |0.3925984  | --- |\n| effnet-b7 |0.3821176  | 0.530|\n| effnetv2-m |0.3869362 | 0.532|\n| swim-transformer |0.3815662  | ---|\n\n***Image Level***\n- YOLOv5 \nyolotrs-384, yolotrl-384,yolov5x-384, yolov5x-512, yolov5x-640.\nI just do yolotrs-384 and yolotrs-384+pseudo with test dataset in image level part.\nyolov5s + transformer  [link](https://www.gitmemory.com/issue/ultralytics/yolov5/2329/808852867)\ndetection mAP@0.5 opacity class.\n\n|  |  yolotrs-384|  yolotrs-384-pseudo |\n| --- | --- | --- |\n| mAP0.5 TTA  | 0.567 | 0.576 |\n| LB  | 0.092| 0.095 |\n\nFor more details, please see https://www.kaggle.com/c/siim-covid19-detection/discussion/264287\n\n***Final Submission***\nFinal Submission Notebook: https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292",
      "votes": null
    },
    {
      "id": "1462959",
      "postDate": "08/10/2021 03:52:29",
      "content": "<p>Congratulations! Thanks for sharing! </p>",
      "rawMarkdown": "Congratulations! Thanks for sharing!",
      "votes": null
    },
    {
      "id": "1464137",
      "postDate": "08/10/2021 13:11:04",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/Roc\" target=\"_blank\">@Roc</a> and thanks for the detailed explanation.</p>",
      "rawMarkdown": "Congratulations @Roc and thanks for the detailed explanation.",
      "votes": null
    },
    {
      "id": "1464374",
      "postDate": "08/10/2021 14:47:58",
      "content": "<p>Congrats <a href=\"https://www.kaggle.com/yingpengchen\" target=\"_blank\">@yingpengchen</a> thanks for sharing</p>",
      "rawMarkdown": "Congrats @yingpengchen thanks for sharing",
      "votes": null
    },
    {
      "id": "1467389",
      "postDate": "08/12/2021 00:44:07",
      "content": "<p>Thank you, looking forward to next time you can get your satisfactory results.</p>",
      "rawMarkdown": "Thank you, looking forward to next time you can get your satisfactory results.",
      "votes": null
    },
    {
      "id": "1467390",
      "postDate": "08/12/2021 00:46:26",
      "content": "<p>You're welcome</p>",
      "rawMarkdown": "You're welcome",
      "votes": null
    },
    {
      "id": "1467391",
      "postDate": "08/12/2021 00:46:32",
      "content": "<p>You're welcome.</p>",
      "rawMarkdown": "You're welcome.",
      "votes": null
    },
    {
      "id": "1468216",
      "postDate": "08/12/2021 09:58:36",
      "content": "<p>Thanks for sharing the code, it's really helpful for people like me who don't know how to get good score! And congratulations! </p>",
      "rawMarkdown": "Thanks for sharing the code, it's really helpful for people like me who don't know how to get good score! And congratulations!",
      "votes": null
    },
    {
      "id": "1468241",
      "postDate": "08/12/2021 10:09:19",
      "content": "<p>You're welcome. Keep kaggling, you will learn more, I also learned in this community.</p>",
      "rawMarkdown": "You're welcome. Keep kaggling, you will learn more, I also learned in this community.",
      "votes": null
    },
    {
      "id": "1469349",
      "postDate": "08/12/2021 21:35:42",
      "content": "<p>Thank you! Could you explain how to make pseudo label with <code>siim covid dataset + test dataset + ricord dataset + bimcv dataset</code>?</p>",
      "rawMarkdown": "Thank you! Could you explain how to make pseudo label with `siim covid dataset + test dataset + ricord dataset + bimcv dataset`?",
      "votes": null
    },
    {
      "id": "1469457",
      "postDate": "08/13/2021 00:28:23",
      "content": "<p>Steps</p>\n<ol>\n<li><p>Training with siim covid dataset.</p></li>\n<li><p>Generate pseudo label with step1 model in test dataset ricord dataset and bimcv dataset.(Note:You need remove duplicate images and invalid images)</p></li>\n</ol>",
      "rawMarkdown": "Steps\n1. Training with siim covid dataset.\n\n1. Generate pseudo label with step1 model in test dataset ricord dataset and bimcv dataset.(Note:You need remove duplicate images and invalid images)",
      "votes": null
    },
    {
      "id": "1469460",
      "postDate": "08/13/2021 00:31:29",
      "content": "<p>Tyank you! I'll try it!</p>",
      "rawMarkdown": "Tyank you! I'll try it!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1462959,
      "author_name": "gordonhu608",
      "author_url": "",
      "post_date": "08/10/2021 03:52:29",
      "content": "<p>Congratulations! Thanks for sharing! </p>",
      "votes": null,
      "replies": [
        {
          "id": 1467389,
          "author_name": "yingpengchen",
          "author_url": "",
          "post_date": "08/12/2021 00:44:07",
          "content": "<p>Thank you, looking forward to next time you can get your satisfactory results.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1464137,
      "author_name": "furcifer",
      "author_url": "",
      "post_date": "08/10/2021 13:11:04",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/Roc\" target=\"_blank\">@Roc</a> and thanks for the detailed explanation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1467390,
          "author_name": "yingpengchen",
          "author_url": "",
          "post_date": "08/12/2021 00:46:26",
          "content": "<p>You're welcome</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1464374,
      "author_name": "jarupula",
      "author_url": "",
      "post_date": "08/10/2021 14:47:58",
      "content": "<p>Congrats <a href=\"https://www.kaggle.com/yingpengchen\" target=\"_blank\">@yingpengchen</a> thanks for sharing</p>",
      "votes": null,
      "replies": [
        {
          "id": 1467391,
          "author_name": "yingpengchen",
          "author_url": "",
          "post_date": "08/12/2021 00:46:32",
          "content": "<p>You're welcome.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1468216,
      "author_name": "liushuzhi",
      "author_url": "",
      "post_date": "08/12/2021 09:58:36",
      "content": "<p>Thanks for sharing the code, it's really helpful for people like me who don't know how to get good score! And congratulations! </p>",
      "votes": null,
      "replies": [
        {
          "id": 1468241,
          "author_name": "yingpengchen",
          "author_url": "",
          "post_date": "08/12/2021 10:09:19",
          "content": "<p>You're welcome. Keep kaggling, you will learn more, I also learned in this community.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1469349,
          "author_name": "liushuzhi",
          "author_url": "",
          "post_date": "08/12/2021 21:35:42",
          "content": "<p>Thank you! Could you explain how to make pseudo label with <code>siim covid dataset + test dataset + ricord dataset + bimcv dataset</code>?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1469457,
          "author_name": "yingpengchen",
          "author_url": "",
          "post_date": "08/13/2021 00:28:23",
          "content": "<p>Steps</p>\n<ol>\n<li><p>Training with siim covid dataset.</p></li>\n<li><p>Generate pseudo label with step1 model in test dataset ricord dataset and bimcv dataset.(Note:You need remove duplicate images and invalid images)</p></li>\n</ol>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1469460,
          "author_name": "liushuzhi",
          "author_url": "",
          "post_date": "08/13/2021 00:31:29",
          "content": "<p>Tyank you! I'll try it!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1462941": "First of all, I would like to thank you for organizing this very interesting competition and congratulations to all the winners of this competition. Also thanks to everyone in our team @quochungto @morizin @joven1997 @socom20 for allowing me to win the first gold medal in this competition. \n\n***Study Level***\n\n- Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) with siim covid dataset .\n- Train classification(4 classes: negative, typical, indeterminate, atypical) + segmentation(boxes to mask) + pseudo label with siim covid dataset + test dataset + ricord dataset + bimcv dataset.（When dealing with bimcv external data, I would like to thank @quochungto and @joven1997 manually deleting the noise data.）\n- Augmentation: Resize, HorizontalFlip, RandomBrightness, RandomContrast, OpticalDistortion, GridDistortion, HueSaturationValue, ShiftScaleRotate, CutoutV2.\n- Loss:  \ncls loss = BiTemperedLogisticLoss  [link](https://github.com/mlpanda/bi-tempered-loss-pytorch/blob/master/bi_tempered_loss.py).\naux loss = 0.9*bce + 0.1*lovasz.\n- Optimizer: Adam with init_learning_rate 0.0001.\n- Scheduler: GradualWarmupSchedulerV2 [link](https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver).\n- Test time augmentation: src+ horizontal flip.\n- Final submission includes 4 models with diversity of encoders, decoders, input size:\nEfficientnet_b5 512, Efficientnet_b7 512, Efficientnetv2_m 512, Swim-Transformer 384.\n- Ensemble method: more details [link](https://www.kaggle.com/cdeotte/forward-selection-oof-ensemble-0-942-private) (Thanks for @morizin lot of ideas)\n\nstudy level part source code used in this competition  has been uploaded to the following github:\nhttps://github.com/ChenYingpeng/pl-siim-covid19-detection\n\n- What didn't work\n     Segmentation channel images as RANZCR competition\n     using more TTA(such as scale=1.25, vertical flip etc)\n     Training 2 classes (none and opacity) in study level.\n     Training  2 classes (none and opacity) in image level. （Use the entire image like none as its box）\n\n arch| cv (5-folds mAP*(4/6))| lb |\n| --- | --- | ---|\n| effnet-b5 |0.387736  | 0.535|\n| effnet-b5-pseudo |0.3925984  | --- |\n| effnet-b7 |0.3821176  | 0.530|\n| effnetv2-m |0.3869362 | 0.532|\n| swim-transformer |0.3815662  | ---|\n\n***Image Level***\n- YOLOv5 \nyolotrs-384, yolotrl-384,yolov5x-384, yolov5x-512, yolov5x-640.\nI just do yolotrs-384 and yolotrs-384+pseudo with test dataset in image level part.\nyolov5s + transformer  [link](https://www.gitmemory.com/issue/ultralytics/yolov5/2329/808852867)\ndetection mAP@0.5 opacity class.\n\n|  |  yolotrs-384|  yolotrs-384-pseudo |\n| --- | --- | --- |\n| mAP0.5 TTA  | 0.567 | 0.576 |\n| LB  | 0.092| 0.095 |\n\nFor more details, please see https://www.kaggle.com/c/siim-covid19-detection/discussion/264287\n\n***Final Submission***\nFinal Submission Notebook: https://www.kaggle.com/yingpengchen/siim-covid19-detection-inference-fast?scriptVersionId=70717292",
    "1462959": "Congratulations! Thanks for sharing!",
    "1464137": "Congratulations @Roc and thanks for the detailed explanation.",
    "1464374": "Congrats @yingpengchen thanks for sharing",
    "1467389": "Thank you, looking forward to next time you can get your satisfactory results.",
    "1467390": "You're welcome",
    "1467391": "You're welcome.",
    "1468216": "Thanks for sharing the code, it's really helpful for people like me who don't know how to get good score! And congratulations!",
    "1468241": "You're welcome. Keep kaggling, you will learn more, I also learned in this community.",
    "1469349": "Thank you! Could you explain how to make pseudo label with `siim covid dataset + test dataset + ricord dataset + bimcv dataset`?",
    "1469457": "Steps\n1. Training with siim covid dataset.\n\n1. Generate pseudo label with step1 model in test dataset ricord dataset and bimcv dataset.(Note:You need remove duplicate images and invalid images)",
    "1469460": "Tyank you! I'll try it!"
  },
  "source": "meta"
}