{
  "id": 392384,
  "title": "ASL Dictionary for making effective features",
  "url": "/competitions/asl-signs/discussion/392384",
  "author_name": "",
  "post_date": "2023-03-05T04:05:26.586419600Z",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>Looking at the actual ASL videos might help making effective features.<br>\nHere is the dictionary.<br>\n<a href=\"https://www.handspeak.com/word/\" target=\"_blank\">https://www.handspeak.com/word/</a></p>\n<p>For example, <a href=\"https://www.handspeak.com/word/727/\" target=\"_blank\">face</a> requires a hand going around the face (see <a href=\"https://www.kaggle.com/code/danielpeshkov/animated-data-visualization\" target=\"_blank\">this notebook</a>), which is hard to recognize when using only mean and std features.<br>\nOn the other hand, making a feature corresponding to cyclic movements (I don't know how. Ideas are welcome!) could help.</p>\n<p>Hope this helps!</p>",
  "messages": [
    {
      "id": "2169372",
      "postDate": "03/05/2023 04:05:26",
      "content": "<p>Hi,</p>\n<p>Looking at the actual ASL videos might help making effective features.<br>\nHere is the dictionary.<br>\n<a href=\"https://www.handspeak.com/word/\" target=\"_blank\">https://www.handspeak.com/word/</a></p>\n<p>For example, <a href=\"https://www.handspeak.com/word/727/\" target=\"_blank\">face</a> requires a hand going around the face (see <a href=\"https://www.kaggle.com/code/danielpeshkov/animated-data-visualization\" target=\"_blank\">this notebook</a>), which is hard to recognize when using only mean and std features.<br>\nOn the other hand, making a feature corresponding to cyclic movements (I don't know how. Ideas are welcome!) could help.</p>\n<p>Hope this helps!</p>",
      "rawMarkdown": "Hi,\n\nLooking at the actual ASL videos might help making effective features.\nHere is the dictionary.\n[https://www.handspeak.com/word/](https://www.handspeak.com/word/)\n\nFor example, [face](https://www.handspeak.com/word/727/) requires a hand going around the face (see [this notebook](https://www.kaggle.com/code/danielpeshkov/animated-data-visualization)), which is hard to recognize when using only mean and std features.\nOn the other hand, making a feature corresponding to cyclic movements (I don't know how. Ideas are welcome!) could help.\n\nHope this helps!",
      "votes": null
    },
    {
      "id": "2169904",
      "postDate": "03/05/2023 14:36:29",
      "content": "<p>i make landmarks visualization  here: <a href=\"https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization\" target=\"_blank\">https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization</a>  please check and i think  <a href=\"https://aslpro.cc\" target=\"_blank\">https://aslpro.cc</a> is better one because it have  dictionary with demo video</p>",
      "rawMarkdown": "i make landmarks visualization  here: https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization  please check and i think  https://aslpro.cc is better one because it have  dictionary with demo video",
      "votes": null
    },
    {
      "id": "2170686",
      "postDate": "03/06/2023 07:46:39",
      "content": "<p>Handspeak also has a <a href=\"https://www.handspeak.com/word/.https://www.handspeak.com/word/\" target=\"_blank\">dictionary with demo videos</a>.</p>",
      "rawMarkdown": "Handspeak also has a [dictionary with demo videos](https://www.handspeak.com/word/.https://www.handspeak.com/word/).",
      "votes": null
    },
    {
      "id": "2170698",
      "postDate": "03/06/2023 07:53:02",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/cnumber\" target=\"_blank\">@cnumber</a>, I was thinking about this too.  I wondered if there were any reverse dictionaries out there, to see how they categorized signs.  (A bit like, for Mandarin, you'd use radical and then stroke count to look up a word.)  I found that Handspeak does have a <a href=\"https://www.handspeak.com/word/asl-eng/\" target=\"_blank\">reverse lookup</a>.  The features they use are…</p>\n<ol>\n<li>Hand shape: One of 27 shapes</li>\n<li>Movement: Unidirectional or repeated</li>\n<li>Location: One of 13 locations (e.g. torso, face, arm, etc.)</li>\n<li>Handed type: (a) one-handed moving, (b) two-handed symmetrical or (c) two-handed alternatively</li>\n</ol>\n<p>And yes, using mean &amp; std doesn't capture this well, which is why we saw such a huge jump in scores with Robert's notebook which included temporal information.</p>",
      "rawMarkdown": "Hi @cnumber, I was thinking about this too.  I wondered if there were any reverse dictionaries out there, to see how they categorized signs.  (A bit like, for Mandarin, you'd use radical and then stroke count to look up a word.)  I found that Handspeak does have a [reverse lookup](https://www.handspeak.com/word/asl-eng/).  The features they use are...\n\n1. Hand shape: One of 27 shapes\n2. Movement: Unidirectional or repeated\n3. Location: One of 13 locations (e.g. torso, face, arm, etc.)\n4. Handed type: (a) one-handed moving, (b) two-handed symmetrical or (c) two-handed alternatively\n\nAnd yes, using mean & std doesn't capture this well, which is why we saw such a huge jump in scores with Robert's notebook which included temporal information.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2169904,
      "author_name": "tantai",
      "author_url": "",
      "post_date": "03/05/2023 14:36:29",
      "content": "<p>i make landmarks visualization  here: <a href=\"https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization\" target=\"_blank\">https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization</a>  please check and i think  <a href=\"https://aslpro.cc\" target=\"_blank\">https://aslpro.cc</a> is better one because it have  dictionary with demo video</p>",
      "votes": null,
      "replies": [
        {
          "id": 2170686,
          "author_name": "andrewrrose",
          "author_url": "",
          "post_date": "03/06/2023 07:46:39",
          "content": "<p>Handspeak also has a <a href=\"https://www.handspeak.com/word/.https://www.handspeak.com/word/\" target=\"_blank\">dictionary with demo videos</a>.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2170698,
      "author_name": "andrewrrose",
      "author_url": "",
      "post_date": "03/06/2023 07:53:02",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/cnumber\" target=\"_blank\">@cnumber</a>, I was thinking about this too.  I wondered if there were any reverse dictionaries out there, to see how they categorized signs.  (A bit like, for Mandarin, you'd use radical and then stroke count to look up a word.)  I found that Handspeak does have a <a href=\"https://www.handspeak.com/word/asl-eng/\" target=\"_blank\">reverse lookup</a>.  The features they use are…</p>\n<ol>\n<li>Hand shape: One of 27 shapes</li>\n<li>Movement: Unidirectional or repeated</li>\n<li>Location: One of 13 locations (e.g. torso, face, arm, etc.)</li>\n<li>Handed type: (a) one-handed moving, (b) two-handed symmetrical or (c) two-handed alternatively</li>\n</ol>\n<p>And yes, using mean &amp; std doesn't capture this well, which is why we saw such a huge jump in scores with Robert's notebook which included temporal information.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2169372": "Hi,\n\nLooking at the actual ASL videos might help making effective features.\nHere is the dictionary.\n[https://www.handspeak.com/word/](https://www.handspeak.com/word/)\n\nFor example, [face](https://www.handspeak.com/word/727/) requires a hand going around the face (see [this notebook](https://www.kaggle.com/code/danielpeshkov/animated-data-visualization)), which is hard to recognize when using only mean and std features.\nOn the other hand, making a feature corresponding to cyclic movements (I don't know how. Ideas are welcome!) could help.\n\nHope this helps!",
    "2169904": "i make landmarks visualization  here: https://www.kaggle.com/code/tantai/gislr-background-knowledge-landmarks-visualization  please check and i think  https://aslpro.cc is better one because it have  dictionary with demo video",
    "2170686": "Handspeak also has a [dictionary with demo videos](https://www.handspeak.com/word/.https://www.handspeak.com/word/).",
    "2170698": "Hi @cnumber, I was thinking about this too.  I wondered if there were any reverse dictionaries out there, to see how they categorized signs.  (A bit like, for Mandarin, you'd use radical and then stroke count to look up a word.)  I found that Handspeak does have a [reverse lookup](https://www.handspeak.com/word/asl-eng/).  The features they use are...\n\n1. Hand shape: One of 27 shapes\n2. Movement: Unidirectional or repeated\n3. Location: One of 13 locations (e.g. torso, face, arm, etc.)\n4. Handed type: (a) one-handed moving, (b) two-handed symmetrical or (c) two-handed alternatively\n\nAnd yes, using mean & std doesn't capture this well, which is why we saw such a huge jump in scores with Robert's notebook which included temporal information."
  },
  "source": "meta"
}