{
  "id": 169804,
  "title": "Yolov5 0.74+ LB reproducibility [learning purposes]",
  "url": "/competitions/global-wheat-detection/discussion/169804",
  "author_name": "Chihaoui007",
  "post_date": "2020-07-25T09:32:01.095000",
  "votes": -2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>As yolov5 is out of the race and for learning purposes, I want to reproduce the public kernels' results using yolov5. Actually when following <a href=\"https://www.kaggle.com/orkatz2/yolov5-fake-or-real-single-model-l-b-0-753\">notebook1</a> and <a href=\"https://www.kaggle.com/nacirbouazizi/well-explained-training-pipeline-to-get-lb-0-760\">notebook2</a>, I only  got around 0.71 LB (single model, no pseudo labeling) . To reproduce the results, I think we need the <strong>exact data split</strong> and which <strong>training options</strong> to activate (ex. --multi-scale, epochs...).\nAlso, I am not using apex but I don't think it has some impact on the model's accuracy.\nSo, can anyone share a reproducible kernel or the full configuration settings?</p>",
  "messages": [
    {
      "id": 944716,
      "postDate": "2020-07-25T09:32:01.097Z",
      "content": "<p>As yolov5 is out of the race and for learning purposes, I want to reproduce the public kernels' results using yolov5. Actually when following <a href=\"https://www.kaggle.com/orkatz2/yolov5-fake-or-real-single-model-l-b-0-753\">notebook1</a> and <a href=\"https://www.kaggle.com/nacirbouazizi/well-explained-training-pipeline-to-get-lb-0-760\">notebook2</a>, I only  got around 0.71 LB (single model, no pseudo labeling) . To reproduce the results, I think we need the <strong>exact data split</strong> and which <strong>training options</strong> to activate (ex. --multi-scale, epochs...).\nAlso, I am not using apex but I don't think it has some impact on the model's accuracy.\nSo, can anyone share a reproducible kernel or the full configuration settings?</p>",
      "rawMarkdown": "As yolov5 is out of the race and for learning purposes, I want to reproduce the public kernels' results using yolov5. Actually when following [notebook1]( https://www.kaggle.com/orkatz2/yolov5-fake-or-real-single-model-l-b-0-753) and [notebook2](https://www.kaggle.com/nacirbouazizi/well-explained-training-pipeline-to-get-lb-0-760), I only  got around 0.71 LB (single model, no pseudo labeling) . To reproduce the results, I think we need the **exact data split** and which **training options** to activate (ex. --multi-scale, epochs...).\nAlso, I am not using apex but I don't think it has some impact on the model's accuracy.\nSo, can anyone share a reproducible kernel or the full configuration settings?",
      "votes": -2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "944716": "As yolov5 is out of the race and for learning purposes, I want to reproduce the public kernels' results using yolov5. Actually when following [notebook1]( https://www.kaggle.com/orkatz2/yolov5-fake-or-real-single-model-l-b-0-753) and [notebook2](https://www.kaggle.com/nacirbouazizi/well-explained-training-pipeline-to-get-lb-0-760), I only  got around 0.71 LB (single model, no pseudo labeling) . To reproduce the results, I think we need the **exact data split** and which **training options** to activate (ex. --multi-scale, epochs...).\nAlso, I am not using apex but I don't think it has some impact on the model's accuracy.\nSo, can anyone share a reproducible kernel or the full configuration settings?"
  }
}