{
  "id": 110355,
  "title": "[private 0.977]: solution",
  "url": "/competitions/recursion-cellular-image-classification/discussion/110355",
  "author_name": "Dake",
  "post_date": "2019-09-27T03:10:25.250000",
  "votes": 8,
  "comment_count": 4,
  "views": 0,
  "content": "<p>our first kaggle competition, and we will keep on:\nmodel：densenet121\nclassification: 1108\ninput: resolution 512*512, rotate, flip, aug(lightness, contrast), normalize...\nensemble: avg</p>\n\n<p>train method: Adam optimizer, decreasing learning rate, iteratively train with tests, \nplate leak, 31controls,  rearrange, </p>\n\n<p>loss: ce, center-loss, arc</p>",
  "messages": [
    {
      "id": 634991,
      "postDate": "2019-09-27T03:10:25.250Z",
      "content": "<p>our first kaggle competition, and we will keep on:\nmodel：densenet121\nclassification: 1108\ninput: resolution 512*512, rotate, flip, aug(lightness, contrast), normalize...\nensemble: avg</p>\n\n<p>train method: Adam optimizer, decreasing learning rate, iteratively train with tests, \nplate leak, 31controls,  rearrange, </p>\n\n<p>loss: ce, center-loss, arc</p>",
      "rawMarkdown": "our first kaggle competition, and we will keep on:\nmodel：densenet121\nclassification: 1108\ninput: resolution 512*512, rotate, flip, aug(lightness, contrast), normalize...\nensemble: avg\n\ntrain method: Adam optimizer, decreasing learning rate, iteratively train with tests, \nplate leak, 31controls,  rearrange, \n\nloss: ce, center-loss, arc",
      "votes": 8
    },
    {
      "id": 635186,
      "postDate": "2019-09-27T08:41:39.347Z",
      "content": "<p><a href=\"/dake0522\">@dake0522</a>  Did you use 6 channel 512*512 for training . Also,does your final solution use cross entropy ,center-loss and arc?</p>",
      "rawMarkdown": "@dake0522  Did you use 6 channel 512*512 for training . Also,does your final solution use cross entropy ,center-loss and arc?",
      "replies": [
        {
          "id": 635692,
          "postDate": "2019-09-28T03:35:01.047Z",
          "content": "<p>At first we tried crop 512 to other resolutions, like 256, 384 etc, not work,  so our final solution is only on 6 channels 512x512 images at both sites. And loss is CE and center-loss.</p>",
          "rawMarkdown": "At first we tried crop 512 to other resolutions, like 256, 384 etc, not work,  so our final solution is only on 6 channels 512x512 images at both sites. And loss is CE and center-loss."
        },
        {
          "id": 636743,
          "postDate": "2019-09-30T06:06:18.653Z",
          "content": "<p>Thanks for the reply.Any plans to release the code? It''ll be really cool to understand the method from implementation perspective.</p>",
          "rawMarkdown": "Thanks for the reply.Any plans to release the code? It''ll be really cool to understand the method from implementation perspective."
        }
      ]
    },
    {
      "id": 635026,
      "postDate": "2019-09-27T04:48:40Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 635186,
      "author_name": "Deepshad",
      "author_url": "",
      "post_date": "2019-09-27T08:41:39.347000",
      "content": "<p><a href=\"/dake0522\">@dake0522</a>  Did you use 6 channel 512*512 for training . Also,does your final solution use cross entropy ,center-loss and arc?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 635692,
          "author_name": "windyfly",
          "author_url": "",
          "post_date": "2019-09-28T03:35:01.047000",
          "content": "<p>At first we tried crop 512 to other resolutions, like 256, 384 etc, not work,  so our final solution is only on 6 channels 512x512 images at both sites. And loss is CE and center-loss.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 636743,
          "author_name": "Deepshad",
          "author_url": "",
          "post_date": "2019-09-30T06:06:18.653000",
          "content": "<p>Thanks for the reply.Any plans to release the code? It''ll be really cool to understand the method from implementation perspective.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 635026,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-09-27T04:48:40",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "634991": "our first kaggle competition, and we will keep on:\nmodel：densenet121\nclassification: 1108\ninput: resolution 512*512, rotate, flip, aug(lightness, contrast), normalize...\nensemble: avg\n\ntrain method: Adam optimizer, decreasing learning rate, iteratively train with tests, \nplate leak, 31controls,  rearrange, \n\nloss: ce, center-loss, arc",
    "635186": "@dake0522  Did you use 6 channel 512*512 for training . Also,does your final solution use cross entropy ,center-loss and arc?",
    "635026": ""
  }
}