{
  "id": 245009,
  "title": "SSD doesn't work.. can you help me?",
  "url": "/competitions/siim-covid19-detection/discussion/245009",
  "author_name": "",
  "post_date": "2021-06-09T09:01:45.643544400Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I'm a beginner. I created an SSD model. As you train, the loss will increase. Something changes, but I don't know what's wrong. can you help me?</p>\n<p><a href=\"https://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training\" target=\"_blank\">https://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training</a></p>",
  "messages": [
    {
      "id": "1342201",
      "postDate": "06/09/2021 09:01:45",
      "content": "<p>I'm a beginner. I created an SSD model. As you train, the loss will increase. Something changes, but I don't know what's wrong. can you help me?</p>\n<p><a href=\"https://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training\" target=\"_blank\">https://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training</a></p>",
      "rawMarkdown": "I'm a beginner. I created an SSD model. As you train, the loss will increase. Something changes, but I don't know what's wrong. can you help me?\n\nhttps://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training",
      "votes": null
    },
    {
      "id": "1342506",
      "postDate": "06/09/2021 14:08:53",
      "content": "<p>to be honest, I didn't understand your code very well, but something caught my eye: for some reason you have 5 classes, although there are only 4 of them in the task - negative = 0 and your learning speed is great, try to reduce it to 1e-4 or even stronger. Check how your bbox changes after resizing - whether you train the mesh at all.</p>",
      "rawMarkdown": "to be honest, I didn't understand your code very well, but something caught my eye: for some reason you have 5 classes, although there are only 4 of them in the task - negative = 0 and your learning speed is great, try to reduce it to 1e-4 or even stronger. Check how your bbox changes after resizing - whether you train the mesh at all.",
      "votes": null
    },
    {
      "id": "1345172",
      "postDate": "06/11/2021 11:20:57",
      "content": "<p>Thank you for looking at the code. I will try what you pointed out immediately.</p>",
      "rawMarkdown": "Thank you for looking at the code. I will try what you pointed out immediately.",
      "votes": null
    },
    {
      "id": "1345206",
      "postDate": "06/11/2021 11:54:40",
      "content": "<p>It seems that the SSD class must be the actual class +1. When I tried it, I got an error.</p>",
      "rawMarkdown": "It seems that the SSD class must be the actual class +1. When I tried it, I got an error.",
      "votes": null
    },
    {
      "id": "1345248",
      "postDate": "06/11/2021 12:21:46",
      "content": "<p>I unfortunately do not know how classes are defined in ssd, but in pytorch 0 is the background class, in this task the negative is the background. therefore, if ssd works in pytorch, then there should be 4 classes.</p>",
      "rawMarkdown": "I unfortunately do not know how classes are defined in ssd, but in pytorch 0 is the background class, in this task the negative is the background. therefore, if ssd works in pytorch, then there should be 4 classes.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1342506,
      "author_name": "zaakciiru",
      "author_url": "",
      "post_date": "06/09/2021 14:08:53",
      "content": "<p>to be honest, I didn't understand your code very well, but something caught my eye: for some reason you have 5 classes, although there are only 4 of them in the task - negative = 0 and your learning speed is great, try to reduce it to 1e-4 or even stronger. Check how your bbox changes after resizing - whether you train the mesh at all.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1345172,
          "author_name": "tensorchoko",
          "author_url": "",
          "post_date": "06/11/2021 11:20:57",
          "content": "<p>Thank you for looking at the code. I will try what you pointed out immediately.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1345206,
          "author_name": "tensorchoko",
          "author_url": "",
          "post_date": "06/11/2021 11:54:40",
          "content": "<p>It seems that the SSD class must be the actual class +1. When I tried it, I got an error.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1345248,
          "author_name": "zaakciiru",
          "author_url": "",
          "post_date": "06/11/2021 12:21:46",
          "content": "<p>I unfortunately do not know how classes are defined in ssd, but in pytorch 0 is the background class, in this task the negative is the background. therefore, if ssd works in pytorch, then there should be 4 classes.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1342201": "I'm a beginner. I created an SSD model. As you train, the loss will increase. Something changes, but I don't know what's wrong. can you help me?\n\nhttps://www.kaggle.com/tensorchoko/siim-pytorch-1-ssd-training",
    "1342506": "to be honest, I didn't understand your code very well, but something caught my eye: for some reason you have 5 classes, although there are only 4 of them in the task - negative = 0 and your learning speed is great, try to reduce it to 1e-4 or even stronger. Check how your bbox changes after resizing - whether you train the mesh at all.",
    "1345172": "Thank you for looking at the code. I will try what you pointed out immediately.",
    "1345206": "It seems that the SSD class must be the actual class +1. When I tried it, I got an error.",
    "1345248": "I unfortunately do not know how classes are defined in ssd, but in pytorch 0 is the background class, in this task the negative is the background. therefore, if ssd works in pytorch, then there should be 4 classes."
  },
  "source": "meta"
}