{
  "id": 309234,
  "title": "Augmenting contrast in images 🤔 ?",
  "url": "/competitions/happy-whale-and-dolphin/discussion/309234",
  "author_name": "",
  "post_date": "2022-02-22T13:39:48.900375900Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi kagglers 👋,<br>\nI've noticed that on some images, there are details on individuals that are almost not noticeable to the naked eye !<br>\nSo I thought why not try to increase the contrast on the images ?</p>\n<p>Here is an example:<br>\n<img src=\"https://i.imgur.com/KBgLGxJ.png\" alt=\"\"><br>\nWe can clearly see that some details start to appear that were kind of \"hidden\".<br>\nAs the values of the pixels are normalized for the models, I wonder if they are able to \"see\" such details or if increasing contrast effectively helps them to learn…</p>\n<p>Anyway, I tried to develop a way of increasing details in images without burning them (i.e. not too much brightness) in my last notebook:<br>\n<a href=\"https://www.kaggle.com/wolfy73/extract-more-information-from-images\" target=\"_blank\">🐳&amp;🐬 - Extract more information from images 🖼️</a><br>\nI will try it on my solution and give feedback.</p>\n<p>Have you tried something along these lines 🤔 ?</p>",
  "messages": [
    {
      "id": "1701045",
      "postDate": "02/22/2022 13:39:48",
      "content": "<p>Hi kagglers 👋,<br>\nI've noticed that on some images, there are details on individuals that are almost not noticeable to the naked eye !<br>\nSo I thought why not try to increase the contrast on the images ?</p>\n<p>Here is an example:<br>\n<img src=\"https://i.imgur.com/KBgLGxJ.png\" alt=\"\"><br>\nWe can clearly see that some details start to appear that were kind of \"hidden\".<br>\nAs the values of the pixels are normalized for the models, I wonder if they are able to \"see\" such details or if increasing contrast effectively helps them to learn…</p>\n<p>Anyway, I tried to develop a way of increasing details in images without burning them (i.e. not too much brightness) in my last notebook:<br>\n<a href=\"https://www.kaggle.com/wolfy73/extract-more-information-from-images\" target=\"_blank\">🐳&amp;🐬 - Extract more information from images 🖼️</a><br>\nI will try it on my solution and give feedback.</p>\n<p>Have you tried something along these lines 🤔 ?</p>",
      "rawMarkdown": "Hi kagglers 👋,\nI've noticed that on some images, there are details on individuals that are almost not noticeable to the naked eye !\nSo I thought why not try to increase the contrast on the images ?\n\nHere is an example:\n![](https://i.imgur.com/KBgLGxJ.png\"/)\nWe can clearly see that some details start to appear that were kind of \"hidden\".\nAs the values of the pixels are normalized for the models, I wonder if they are able to \"see\" such details or if increasing contrast effectively helps them to learn...\n\nAnyway, I tried to develop a way of increasing details in images without burning them (i.e. not too much brightness) in my last notebook:\n[🐳&🐬 - Extract more information from images 🖼️](https://www.kaggle.com/wolfy73/extract-more-information-from-images)\nI will try it on my solution and give feedback.\n\nHave you tried something along these lines 🤔 ?",
      "votes": null
    },
    {
      "id": "1701067",
      "postDate": "02/22/2022 13:54:55",
      "content": "<p>Attitude for computer vision problems - data are for computer (model) not for people. Sometimes better to give convolution to extract patterns than manipulate to see (human) as better image. When you create such experiment feed NN and check if this works. Sometimes works, sometimes does not …. but NN should eveluate this. </p>\n<p>In this example you clipped information from \"lights\" area. You can see some burned out spots. You lost information unfortunately. In contrast you can loose information in shadow area as well. </p>",
      "rawMarkdown": "Attitude for computer vision problems - data are for computer (model) not for people. Sometimes better to give convolution to extract patterns than manipulate to see (human) as better image. When you create such experiment feed NN and check if this works. Sometimes works, sometimes does not .... but NN should eveluate this. \n\nIn this example you clipped information from \"lights\" area. You can see some burned out spots. You lost information unfortunately. In contrast you can loose information in shadow area as well.",
      "votes": null
    },
    {
      "id": "1701280",
      "postDate": "02/22/2022 16:19:01",
      "content": "<p>Very wise advice !</p>",
      "rawMarkdown": "Very wise advice !",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1701067,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "02/22/2022 13:54:55",
      "content": "<p>Attitude for computer vision problems - data are for computer (model) not for people. Sometimes better to give convolution to extract patterns than manipulate to see (human) as better image. When you create such experiment feed NN and check if this works. Sometimes works, sometimes does not …. but NN should eveluate this. </p>\n<p>In this example you clipped information from \"lights\" area. You can see some burned out spots. You lost information unfortunately. In contrast you can loose information in shadow area as well. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1701280,
          "author_name": "wolfy73",
          "author_url": "",
          "post_date": "02/22/2022 16:19:01",
          "content": "<p>Very wise advice !</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1701045": "Hi kagglers 👋,\nI've noticed that on some images, there are details on individuals that are almost not noticeable to the naked eye !\nSo I thought why not try to increase the contrast on the images ?\n\nHere is an example:\n![](https://i.imgur.com/KBgLGxJ.png\"/)\nWe can clearly see that some details start to appear that were kind of \"hidden\".\nAs the values of the pixels are normalized for the models, I wonder if they are able to \"see\" such details or if increasing contrast effectively helps them to learn...\n\nAnyway, I tried to develop a way of increasing details in images without burning them (i.e. not too much brightness) in my last notebook:\n[🐳&🐬 - Extract more information from images 🖼️](https://www.kaggle.com/wolfy73/extract-more-information-from-images)\nI will try it on my solution and give feedback.\n\nHave you tried something along these lines 🤔 ?",
    "1701067": "Attitude for computer vision problems - data are for computer (model) not for people. Sometimes better to give convolution to extract patterns than manipulate to see (human) as better image. When you create such experiment feed NN and check if this works. Sometimes works, sometimes does not .... but NN should eveluate this. \n\nIn this example you clipped information from \"lights\" area. You can see some burned out spots. You lost information unfortunately. In contrast you can loose information in shadow area as well.",
    "1701280": "Very wise advice !"
  },
  "source": "meta"
}