{
  "id": 413347,
  "title": "Beginners Problem: Can anyone help me to create a dataset object (Pytorch)",
  "url": "/competitions/asl-fingerspelling/discussion/413347",
  "author_name": "smittenk",
  "post_date": "2023-05-28T06:45:13.151000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hey,<br>\nI am total beginner in deep learning and I want your help to create a dataset without any transformation or augmentation.<br>\nI tried doing the task but due to the different #frames per video is making me a problem to create a dataset.</p>",
  "messages": [
    {
      "id": 2279779,
      "postDate": "2023-05-29T16:11:25.040Z",
      "content": "<p>You could try pad_sequence, refer to <a href=\"https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence\" target=\"_blank\">https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence</a></p>",
      "rawMarkdown": "You could try pad_sequence, refer to https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence",
      "replies": [
        {
          "id": 2280694,
          "postDate": "2023-05-30T09:53:45.657Z",
          "content": "<p><a href=\"https://www.kaggle.com/jiayezou\" target=\"_blank\">@jiayezou</a> thank you for this.</p>",
          "rawMarkdown": "@jiayezou thank you for this.\n"
        },
        {
          "id": 2280700,
          "postDate": "2023-05-30T09:59:01.867Z",
          "content": "<p>According to the competition guidelines, the model should predict a character or null value for each frame. However, I am facing a challenge in mapping the input landmark coordinates to a single character when the true label character is not provided for each frame. Consequently, calculating the loss becomes problematic without the true labels at each frame.</p>\n<p>Could you provide any guidance or insights on how to approach this matter ?</p>",
          "rawMarkdown": "According to the competition guidelines, the model should predict a character or null value for each frame. However, I am facing a challenge in mapping the input landmark coordinates to a single character when the true label character is not provided for each frame. Consequently, calculating the loss becomes problematic without the true labels at each frame.\n\nCould you provide any guidance or insights on how to approach this matter ?"
        }
      ]
    },
    {
      "id": 2277898,
      "postDate": "2023-05-28T06:45:13.150Z",
      "content": "<p>Hey,<br>\nI am total beginner in deep learning and I want your help to create a dataset without any transformation or augmentation.<br>\nI tried doing the task but due to the different #frames per video is making me a problem to create a dataset.</p>",
      "rawMarkdown": "Hey,\nI am total beginner in deep learning and I want your help to create a dataset without any transformation or augmentation.\nI tried doing the task but due to the different #frames per video is making me a problem to create a dataset.\n\n"
    }
  ],
  "comments": [
    {
      "id": 2279779,
      "author_name": "Jiaye Zou",
      "author_url": "",
      "post_date": "2023-05-29T16:11:25.040000",
      "content": "<p>You could try pad_sequence, refer to <a href=\"https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence\" target=\"_blank\">https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence</a></p>",
      "votes": 0,
      "replies": [
        {
          "id": 2280694,
          "author_name": "smittenk",
          "author_url": "",
          "post_date": "2023-05-30T09:53:45.657000",
          "content": "<p><a href=\"https://www.kaggle.com/jiayezou\" target=\"_blank\">@jiayezou</a> thank you for this.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 2280700,
          "author_name": "smittenk",
          "author_url": "",
          "post_date": "2023-05-30T09:59:01.867000",
          "content": "<p>According to the competition guidelines, the model should predict a character or null value for each frame. However, I am facing a challenge in mapping the input landmark coordinates to a single character when the true label character is not provided for each frame. Consequently, calculating the loss becomes problematic without the true labels at each frame.</p>\n<p>Could you provide any guidance or insights on how to approach this matter ?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2279779": "You could try pad_sequence, refer to https://pytorch.org/docs/stable/generated/torch.nn.utils.rnn.pad_sequence.html#torch.nn.utils.rnn.pad_sequence",
    "2277898": "Hey,\nI am total beginner in deep learning and I want your help to create a dataset without any transformation or augmentation.\nI tried doing the task but due to the different #frames per video is making me a problem to create a dataset.\n\n"
  }
}