{
  "id": 105970,
  "title": "Convergence of YOLOv2 using Darkflow",
  "url": "/competitions/kuzushiji-recognition/discussion/105970",
  "author_name": "Matteo Rizzo",
  "post_date": "2019-08-27T12:54:29.931000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi everyone! Was anybody able to obtain reasonable results using YOLOv2 with Darkflow? I've tried really a lot of different network and training parameters configurations but I've not been able to diminish the loss under 30/40, nor to get decent bounding boxes even using a small subset of the data to test the model and try to overfit. Any help or observation would be much appreciated, thanks.</p>",
  "messages": [
    {
      "id": 609151,
      "postDate": "2019-08-27T12:54:29.933Z",
      "content": "<p>Hi everyone! Was anybody able to obtain reasonable results using YOLOv2 with Darkflow? I've tried really a lot of different network and training parameters configurations but I've not been able to diminish the loss under 30/40, nor to get decent bounding boxes even using a small subset of the data to test the model and try to overfit. Any help or observation would be much appreciated, thanks.</p>",
      "rawMarkdown": "Hi everyone! Was anybody able to obtain reasonable results using YOLOv2 with Darkflow? I've tried really a lot of different network and training parameters configurations but I've not been able to diminish the loss under 30/40, nor to get decent bounding boxes even using a small subset of the data to test the model and try to overfit. Any help or observation would be much appreciated, thanks.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "609151": "Hi everyone! Was anybody able to obtain reasonable results using YOLOv2 with Darkflow? I've tried really a lot of different network and training parameters configurations but I've not been able to diminish the loss under 30/40, nor to get decent bounding boxes even using a small subset of the data to test the model and try to overfit. Any help or observation would be much appreciated, thanks."
  }
}