{
  "id": 154983,
  "title": "[COCO v.s. ImageNet v.s. Others] What Pretrained Backbone You Use?",
  "url": "/competitions/global-wheat-detection/discussion/154983",
  "author_name": "",
  "post_date": "2020-05-30T17:49:30.677017400Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I wanna see a voting/ brief statistics, did you use COCO pretrained model or ImageNet pretrained model or others? Did anyone try different pretrained backbones and see a big performance difference? Would like to see your sharing.</p>\n\n<p>For me, I am using ImageNet pretrained backbone. The performance is not good at all. (It may or may not be related the pretrained backbone, it's sth I have to figure out)</p>",
  "messages": [
    {
      "id": "867884",
      "postDate": "05/30/2020 17:49:30",
      "content": "<p>I wanna see a voting/ brief statistics, did you use COCO pretrained model or ImageNet pretrained model or others? Did anyone try different pretrained backbones and see a big performance difference? Would like to see your sharing.</p>\n\n<p>For me, I am using ImageNet pretrained backbone. The performance is not good at all. (It may or may not be related the pretrained backbone, it's sth I have to figure out)</p>",
      "rawMarkdown": "I wanna see a voting/ brief statistics, did you use COCO pretrained model or ImageNet pretrained model or others? Did anyone try different pretrained backbones and see a big performance difference? Would like to see your sharing.\n\nFor me, I am using ImageNet pretrained backbone. The performance is not good at all. (It may or may not be related the pretrained backbone, it's sth I have to figure out)",
      "votes": null
    },
    {
      "id": "877449",
      "postDate": "06/07/2020 15:58:23",
      "content": "<p>Probably you can try COCO. </p>\n\n<p>Look at this notebook <a href=\"https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\">https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train</a></p>\n\n<p>I think I will be using COCO as well.</p>",
      "rawMarkdown": "Probably you can try COCO. \n\nLook at this notebook https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\n\nI think I will be using COCO as well.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 877449,
      "author_name": "luisvivas",
      "author_url": "",
      "post_date": "06/07/2020 15:58:23",
      "content": "<p>Probably you can try COCO. </p>\n\n<p>Look at this notebook <a href=\"https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\">https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train</a></p>\n\n<p>I think I will be using COCO as well.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "867884": "I wanna see a voting/ brief statistics, did you use COCO pretrained model or ImageNet pretrained model or others? Did anyone try different pretrained backbones and see a big performance difference? Would like to see your sharing.\n\nFor me, I am using ImageNet pretrained backbone. The performance is not good at all. (It may or may not be related the pretrained backbone, it's sth I have to figure out)",
    "877449": "Probably you can try COCO. \n\nLook at this notebook https://www.kaggle.com/pestipeti/pytorch-starter-fasterrcnn-train\n\nI think I will be using COCO as well."
  },
  "source": "meta"
}