{
  "id": 236017,
  "title": "better encoder?",
  "url": "/competitions/bms-molecular-translation/discussion/236017",
  "author_name": "",
  "post_date": "2021-05-02T12:39:23.597862100Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I found that almost all the public encoder is based on efficient-net. <br>\nIs it nice to change the other encoder like resnet or whatever it is.<br>\nI'm trying to pretrain the model by generating a picture that is similar to the input. Then save the weight , change the encoder, and train on the dataset.</p>",
  "messages": [
    {
      "id": "1290810",
      "postDate": "05/02/2021 12:39:23",
      "content": "<p>I found that almost all the public encoder is based on efficient-net. <br>\nIs it nice to change the other encoder like resnet or whatever it is.<br>\nI'm trying to pretrain the model by generating a picture that is similar to the input. Then save the weight , change the encoder, and train on the dataset.</p>",
      "rawMarkdown": "I found that almost all the public encoder is based on efficient-net. \nIs it nice to change the other encoder like resnet or whatever it is.\nI'm trying to pretrain the model by generating a picture that is similar to the input. Then save the weight , change the encoder, and train on the dataset.",
      "votes": null
    },
    {
      "id": "1291159",
      "postDate": "05/02/2021 18:29:31",
      "content": "<p>I think efficient-net or other large architecture that performs well on imagenet has ability to extract better features from chemical formulas =) therefore A lot of people are using them. </p>",
      "rawMarkdown": "I think efficient-net or other large architecture that performs well on imagenet has ability to extract better features from chemical formulas =) therefore A lot of people are using them.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1291159,
      "author_name": "drhabib",
      "author_url": "",
      "post_date": "05/02/2021 18:29:31",
      "content": "<p>I think efficient-net or other large architecture that performs well on imagenet has ability to extract better features from chemical formulas =) therefore A lot of people are using them. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1290810": "I found that almost all the public encoder is based on efficient-net. \nIs it nice to change the other encoder like resnet or whatever it is.\nI'm trying to pretrain the model by generating a picture that is similar to the input. Then save the weight , change the encoder, and train on the dataset.",
    "1291159": "I think efficient-net or other large architecture that performs well on imagenet has ability to extract better features from chemical formulas =) therefore A lot of people are using them."
  },
  "source": "meta"
}