{
  "id": 95042,
  "title": "post your funniest misclassifications",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/95042",
  "author_name": "",
  "post_date": "2019-06-09T14:05:46.812067100Z",
  "votes": 1,
  "comment_count": 7,
  "views": 0,
  "content": "<p><img src=\"https://imgur.com/6KVUV12\" alt=\"mmmm, sleeve juice\"></p>\n\n<p>also, why so many bags?</p>",
  "messages": [
    {
      "id": "548515",
      "postDate": "06/09/2019 14:05:46",
      "content": "<p><img src=\"https://imgur.com/6KVUV12\" alt=\"mmmm, sleeve juice\"></p>\n\n<p>also, why so many bags?</p>",
      "rawMarkdown": "![mmmm, sleeve juice](https://imgur.com/6KVUV12)\n\nalso, why so many bags?",
      "votes": null
    },
    {
      "id": "548807",
      "postDate": "06/09/2019 23:52:49",
      "content": "<p><img src=\"https://i.ibb.co/7WFr93g/13656.png\" alt=\"\"></p>\n\n<p>My model thinks that this is a dress.</p>",
      "rawMarkdown": "![](https://i.ibb.co/7WFr93g/13656.png)\n\nMy model thinks that this is a dress.",
      "votes": null
    },
    {
      "id": "548852",
      "postDate": "06/10/2019 02:43:40",
      "content": "<p>I love it. Someday I will figure out how to embed images in forum posts.</p>",
      "rawMarkdown": "I love it. Someday I will figure out how to embed images in forum posts.",
      "votes": null
    },
    {
      "id": "549249",
      "postDate": "06/10/2019 13:08:47",
      "content": "<p>Three shoe-like bottles on the table!\n<a href=\"https://storage.googleapis.com/kaggle-forum-message-attachments/549249/13460/image.png\"></a></p>",
      "rawMarkdown": "Three shoe-like bottles on the table!\n[](https://storage.googleapis.com/kaggle-forum-message-attachments/549249/13460/image.png)",
      "votes": null
    },
    {
      "id": "549288",
      "postDate": "06/10/2019 14:06:46",
      "content": "<p>I suppose cups and bottles are more shoe-shaped than anything else. Though I think these errors are quite fascinating as they reveal something about what the models are paying attention to. </p>\n\n<p>I’m hoping I learn enough from doing the new visual relation challenge to be able to explicitly instruct the models to attend to things like relative scale and likely and unlikely spatial proximities. </p>\n\n<p>I mean a bag or purse could actually occupy more pixels in an image than a person if it’s much closer to the camera, and a bunch of shoes can be on a table rather than on feet, but separating everything will need local and non-local attention and probably some monocular depth estimation also. Super interesting. </p>\n\n<p>Thanks for posting!</p>",
      "rawMarkdown": "I suppose cups and bottles are more shoe-shaped than anything else. Though I think these errors are quite fascinating as they reveal something about what the models are paying attention to. \n\nI’m hoping I learn enough from doing the new visual relation challenge to be able to explicitly instruct the models to attend to things like relative scale and likely and unlikely spatial proximities. \n\nI mean a bag or purse could actually occupy more pixels in an image than a person if it’s much closer to the camera, and a bunch of shoes can be on a table rather than on feet, but separating everything will need local and non-local attention and probably some monocular depth estimation also. Super interesting. \n\nThanks for posting!",
      "votes": null
    },
    {
      "id": "549650",
      "postDate": "06/10/2019 22:46:02",
      "content": "<p>I got that one too.</p>",
      "rawMarkdown": "I got that one too.",
      "votes": null
    },
    {
      "id": "549705",
      "postDate": "06/11/2019 00:26:13",
      "content": "<p>Got a similar one too! For me, part of the problem was that I initially forgot to change (or get rid of) the random seed for sampling when iteratively training with my Kaggle kernel, so the NN was apparently seeing a lot of high heels...</p>",
      "rawMarkdown": "Got a similar one too! For me, part of the problem was that I initially forgot to change (or get rid of) the random seed for sampling when iteratively training with my Kaggle kernel, so the NN was apparently seeing a lot of high heels...",
      "votes": null
    },
    {
      "id": "549709",
      "postDate": "06/11/2019 00:29:13",
      "content": "<p>Here you go - my model was confused about hair vs hat, eyes vs glasses, pants vs dogs lol</p>",
      "rawMarkdown": "Here you go - my model was confused about hair vs hat, eyes vs glasses, pants vs dogs lol",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 548807,
      "author_name": "alexanderliao",
      "author_url": "",
      "post_date": "06/09/2019 23:52:49",
      "content": "<p><img src=\"https://i.ibb.co/7WFr93g/13656.png\" alt=\"\"></p>\n\n<p>My model thinks that this is a dress.</p>",
      "votes": null,
      "replies": [
        {
          "id": 548852,
          "author_name": "interneuron",
          "author_url": "",
          "post_date": "06/10/2019 02:43:40",
          "content": "<p>I love it. Someday I will figure out how to embed images in forum posts.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 549249,
      "author_name": "micajoumathematics",
      "author_url": "",
      "post_date": "06/10/2019 13:08:47",
      "content": "<p>Three shoe-like bottles on the table!\n<a href=\"https://storage.googleapis.com/kaggle-forum-message-attachments/549249/13460/image.png\"></a></p>",
      "votes": null,
      "replies": [
        {
          "id": 549288,
          "author_name": "interneuron",
          "author_url": "",
          "post_date": "06/10/2019 14:06:46",
          "content": "<p>I suppose cups and bottles are more shoe-shaped than anything else. Though I think these errors are quite fascinating as they reveal something about what the models are paying attention to. </p>\n\n<p>I’m hoping I learn enough from doing the new visual relation challenge to be able to explicitly instruct the models to attend to things like relative scale and likely and unlikely spatial proximities. </p>\n\n<p>I mean a bag or purse could actually occupy more pixels in an image than a person if it’s much closer to the camera, and a bunch of shoes can be on a table rather than on feet, but separating everything will need local and non-local attention and probably some monocular depth estimation also. Super interesting. </p>\n\n<p>Thanks for posting!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 549650,
          "author_name": "alexanderliao",
          "author_url": "",
          "post_date": "06/10/2019 22:46:02",
          "content": "<p>I got that one too.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 549705,
          "author_name": "enwei26",
          "author_url": "",
          "post_date": "06/11/2019 00:26:13",
          "content": "<p>Got a similar one too! For me, part of the problem was that I initially forgot to change (or get rid of) the random seed for sampling when iteratively training with my Kaggle kernel, so the NN was apparently seeing a lot of high heels...</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 549709,
      "author_name": "enwei26",
      "author_url": "",
      "post_date": "06/11/2019 00:29:13",
      "content": "<p>Here you go - my model was confused about hair vs hat, eyes vs glasses, pants vs dogs lol</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "548515": "![mmmm, sleeve juice](https://imgur.com/6KVUV12)\n\nalso, why so many bags?",
    "548807": "![](https://i.ibb.co/7WFr93g/13656.png)\n\nMy model thinks that this is a dress.",
    "548852": "I love it. Someday I will figure out how to embed images in forum posts.",
    "549249": "Three shoe-like bottles on the table!\n[](https://storage.googleapis.com/kaggle-forum-message-attachments/549249/13460/image.png)",
    "549288": "I suppose cups and bottles are more shoe-shaped than anything else. Though I think these errors are quite fascinating as they reveal something about what the models are paying attention to. \n\nI’m hoping I learn enough from doing the new visual relation challenge to be able to explicitly instruct the models to attend to things like relative scale and likely and unlikely spatial proximities. \n\nI mean a bag or purse could actually occupy more pixels in an image than a person if it’s much closer to the camera, and a bunch of shoes can be on a table rather than on feet, but separating everything will need local and non-local attention and probably some monocular depth estimation also. Super interesting. \n\nThanks for posting!",
    "549650": "I got that one too.",
    "549705": "Got a similar one too! For me, part of the problem was that I initially forgot to change (or get rid of) the random seed for sampling when iteratively training with my Kaggle kernel, so the NN was apparently seeing a lot of high heels...",
    "549709": "Here you go - my model was confused about hair vs hat, eyes vs glasses, pants vs dogs lol"
  },
  "source": "meta"
}