{
  "id": 131985,
  "title": "I want to know whether I have some ways to use pre-trained model for SE ResNest in pytorch",
  "url": "/competitions/bengaliai-cv19/discussion/131985",
  "author_name": "",
  "post_date": "2020-02-23T05:26:33.676204Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>As a new hand,I need your help.I find there are some pre-trained model in pytorch,such as ResNet.But now I need to use SE ResNext.I want to use pre-trained model,but I can't find.\nI wonder if there is a SE ResNext pre-trained model.And how I need to do to use SE ResNext's pre-trained model.</p>",
  "messages": [
    {
      "id": "754141",
      "postDate": "02/23/2020 05:26:33",
      "content": "<p>As a new hand,I need your help.I find there are some pre-trained model in pytorch,such as ResNet.But now I need to use SE ResNext.I want to use pre-trained model,but I can't find.\nI wonder if there is a SE ResNext pre-trained model.And how I need to do to use SE ResNext's pre-trained model.</p>",
      "rawMarkdown": "As a new hand,I need your help.I find there are some pre-trained model in pytorch,such as ResNet.But now I need to use SE ResNext.I want to use pre-trained model,but I can't find.\nI wonder if there is a SE ResNext pre-trained model.And how I need to do to use SE ResNext's pre-trained model.",
      "votes": null
    },
    {
      "id": "754144",
      "postDate": "02/23/2020 05:33:00",
      "content": "<p>it is simple,for example you can use this public kernel : <a href=\"https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654\">https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654</a></p>\n\n<p>and check 4th cell there,  if you replace : </p>\n\n<p>self.backbone = torchvision.models.resnet18(pretrained=False)\nto\nself.backbone = torchvision.models.resnext50_32x4d(pretrained=True)</p>\n\n<p>and delete this line of code : \nself.backbone.load_state_dict(torch.load(\"/kaggle/input/resnet18/resnet18.pth\"))</p>\n\n<p>then using internet you can get the pretrained model resnext50_32x4d \ni hope it answers your question,thanks</p>",
      "rawMarkdown": "it is simple,for example you can use this public kernel : https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654\n\nand check 4th cell there,  if you replace : \n\nself.backbone = torchvision.models.resnet18(pretrained=False)\nto\nself.backbone = torchvision.models.resnext50_32x4d(pretrained=True)\n\nand delete this line of code : \nself.backbone.load_state_dict(torch.load(\"/kaggle/input/resnet18/resnet18.pth\"))\n\nthen using internet you can get the pretrained model resnext50_32x4d \ni hope it answers your question,thanks",
      "votes": null
    },
    {
      "id": "754146",
      "postDate": "02/23/2020 05:37:38",
      "content": "<p>thanks for your help，I look it up for a long time</p>",
      "rawMarkdown": "thanks for your help，I look it up for a long time",
      "votes": null
    },
    {
      "id": "754166",
      "postDate": "02/23/2020 06:31:01",
      "content": "<p>Coincidentally, I just came back from a similar search in a Kaggle discussion that took me to github and recalled that I saw SE resnext in the list. Try this --  <a href=\"https://github.com/Cadene/pretrained-models.pytorch#pnasnet\">https://github.com/Cadene/pretrained-models.pytorch#pnasnet</a> </p>",
      "rawMarkdown": "Coincidentally, I just came back from a similar search in a Kaggle discussion that took me to github and recalled that I saw SE resnext in the list. Try this --  https://github.com/Cadene/pretrained-models.pytorch#pnasnet",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 754144,
      "author_name": "mobassir",
      "author_url": "",
      "post_date": "02/23/2020 05:33:00",
      "content": "<p>it is simple,for example you can use this public kernel : <a href=\"https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654\">https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654</a></p>\n\n<p>and check 4th cell there,  if you replace : </p>\n\n<p>self.backbone = torchvision.models.resnet18(pretrained=False)\nto\nself.backbone = torchvision.models.resnext50_32x4d(pretrained=True)</p>\n\n<p>and delete this line of code : \nself.backbone.load_state_dict(torch.load(\"/kaggle/input/resnet18/resnet18.pth\"))</p>\n\n<p>then using internet you can get the pretrained model resnext50_32x4d \ni hope it answers your question,thanks</p>",
      "votes": null,
      "replies": [
        {
          "id": 754146,
          "author_name": "apstars",
          "author_url": "",
          "post_date": "02/23/2020 05:37:38",
          "content": "<p>thanks for your help，I look it up for a long time</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 754166,
      "author_name": "practicalvisionary",
      "author_url": "",
      "post_date": "02/23/2020 06:31:01",
      "content": "<p>Coincidentally, I just came back from a similar search in a Kaggle discussion that took me to github and recalled that I saw SE resnext in the list. Try this --  <a href=\"https://github.com/Cadene/pretrained-models.pytorch#pnasnet\">https://github.com/Cadene/pretrained-models.pytorch#pnasnet</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "754141": "As a new hand,I need your help.I find there are some pre-trained model in pytorch,such as ResNet.But now I need to use SE ResNext.I want to use pre-trained model,but I can't find.\nI wonder if there is a SE ResNext pre-trained model.And how I need to do to use SE ResNext's pre-trained model.",
    "754144": "it is simple,for example you can use this public kernel : https://www.kaggle.com/pestipeti/simple-pytorch-training-lb-0-7654\n\nand check 4th cell there,  if you replace : \n\nself.backbone = torchvision.models.resnet18(pretrained=False)\nto\nself.backbone = torchvision.models.resnext50_32x4d(pretrained=True)\n\nand delete this line of code : \nself.backbone.load_state_dict(torch.load(\"/kaggle/input/resnet18/resnet18.pth\"))\n\nthen using internet you can get the pretrained model resnext50_32x4d \ni hope it answers your question,thanks",
    "754146": "thanks for your help，I look it up for a long time",
    "754166": "Coincidentally, I just came back from a similar search in a Kaggle discussion that took me to github and recalled that I saw SE resnext in the list. Try this --  https://github.com/Cadene/pretrained-models.pytorch#pnasnet"
  },
  "source": "meta"
}