{
  "id": 170283,
  "title": "Does COCO dataset and COCO based pretrained models allowed?",
  "url": "/competitions/global-wheat-detection/discussion/170283",
  "author_name": "nvnn",
  "post_date": "2020-07-27T05:57:17.384000",
  "votes": 8,
  "comment_count": 4,
  "views": 0,
  "content": "<p>According to the COCO website, the project does not own copyright to images. Additionally, all the images I checked appear to have been copied in violation of their original license. All images are currently being hosted illegally by Google, and even though the cocoapi is BSD licensed, this license does not cover the dataset itself.\nThen, all the pretrained models using COCO dataset will not be allowed. FasterRCNN pretrained models, EfficinetDet pretrained models, etc will not meet MIT licence. \nImagenet is only for non-commercial uses also.\nAny comments on this?</p>",
  "messages": [
    {
      "id": 947189,
      "postDate": "2020-07-27T05:57:17.383Z",
      "content": "<p>According to the COCO website, the project does not own copyright to images. Additionally, all the images I checked appear to have been copied in violation of their original license. All images are currently being hosted illegally by Google, and even though the cocoapi is BSD licensed, this license does not cover the dataset itself.\nThen, all the pretrained models using COCO dataset will not be allowed. FasterRCNN pretrained models, EfficinetDet pretrained models, etc will not meet MIT licence. \nImagenet is only for non-commercial uses also.\nAny comments on this?</p>",
      "rawMarkdown": "According to the COCO website, the project does not own copyright to images. Additionally, all the images I checked appear to have been copied in violation of their original license. All images are currently being hosted illegally by Google, and even though the cocoapi is BSD licensed, this license does not cover the dataset itself.\nThen, all the pretrained models using COCO dataset will not be allowed. FasterRCNN pretrained models, EfficinetDet pretrained models, etc will not meet MIT licence. \nImagenet is only for non-commercial uses also.\nAny comments on this?",
      "votes": 8
    },
    {
      "id": 948659,
      "postDate": "2020-07-28T06:15:48.073Z",
      "content": "<p>I'm not a lawyer but my understanding is that copyright is intended to cover who can <em>copy</em> and distribute the material. E.g. you can't legally sell copies of Harry Potter unless you're J.K.Rowling. Since the images can't be reconstructed from the COCO weights, it doesn't seem like they would count as a violation of the owner's copyright. I imagine this is one of those grey areas that lawyers could argue over.</p>\n\n<p>The copy-left provision in the GPL license (YoLoV5) is pretty explicitly incompatible with the MIT License which is why it was banned. Again, I'm not a lawyer.</p>",
      "rawMarkdown": "I'm not a lawyer but my understanding is that copyright is intended to cover who can *copy* and distribute the material. E.g. you can't legally sell copies of Harry Potter unless you're J.K.Rowling. Since the images can't be reconstructed from the COCO weights, it doesn't seem like they would count as a violation of the owner's copyright. I imagine this is one of those grey areas that lawyers could argue over.\n\nThe copy-left provision in the GPL license (YoLoV5) is pretty explicitly incompatible with the MIT License which is why it was banned. Again, I'm not a lawyer.\n\n\n",
      "votes": 3
    },
    {
      "id": 948014,
      "postDate": "2020-07-27T15:59:53.383Z",
      "content": "<p>The deadline for the competition is comming but I am still confused. 😨 </p>\n\n<p>It doesn't seem to be just a problem with yolov5.</p>",
      "rawMarkdown": "The deadline for the competition is comming but I am still confused. 😨 \n\nIt doesn't seem to be just a problem with yolov5.",
      "votes": 2
    },
    {
      "id": 947219,
      "postDate": "2020-07-27T06:08:57.100Z",
      "rawMarkdown": "",
      "votes": -7,
      "isDeleted": true,
      "replies": [
        {
          "id": 947827,
          "postDate": "2020-07-27T14:00:10.723Z",
          "content": "<p>I think external data must meet the MIT license?</p>",
          "rawMarkdown": "I think external data must meet the MIT license?",
          "votes": -1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 948659,
      "author_name": "Ryan Epp",
      "author_url": "",
      "post_date": "2020-07-28T06:15:48.073000",
      "content": "<p>I'm not a lawyer but my understanding is that copyright is intended to cover who can <em>copy</em> and distribute the material. E.g. you can't legally sell copies of Harry Potter unless you're J.K.Rowling. Since the images can't be reconstructed from the COCO weights, it doesn't seem like they would count as a violation of the owner's copyright. I imagine this is one of those grey areas that lawyers could argue over.</p>\n\n<p>The copy-left provision in the GPL license (YoLoV5) is pretty explicitly incompatible with the MIT License which is why it was banned. Again, I'm not a lawyer.</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 948014,
      "author_name": "Heroseo",
      "author_url": "",
      "post_date": "2020-07-27T15:59:53.383000",
      "content": "<p>The deadline for the competition is comming but I am still confused. 😨 </p>\n\n<p>It doesn't seem to be just a problem with yolov5.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 947219,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-07-27T06:08:57.100000",
      "content": "",
      "votes": -7,
      "replies": [
        {
          "id": 947827,
          "author_name": "nvnn",
          "author_url": "",
          "post_date": "2020-07-27T14:00:10.723000",
          "content": "<p>I think external data must meet the MIT license?</p>",
          "votes": -1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "947189": "According to the COCO website, the project does not own copyright to images. Additionally, all the images I checked appear to have been copied in violation of their original license. All images are currently being hosted illegally by Google, and even though the cocoapi is BSD licensed, this license does not cover the dataset itself.\nThen, all the pretrained models using COCO dataset will not be allowed. FasterRCNN pretrained models, EfficinetDet pretrained models, etc will not meet MIT licence. \nImagenet is only for non-commercial uses also.\nAny comments on this?",
    "948659": "I'm not a lawyer but my understanding is that copyright is intended to cover who can *copy* and distribute the material. E.g. you can't legally sell copies of Harry Potter unless you're J.K.Rowling. Since the images can't be reconstructed from the COCO weights, it doesn't seem like they would count as a violation of the owner's copyright. I imagine this is one of those grey areas that lawyers could argue over.\n\nThe copy-left provision in the GPL license (YoLoV5) is pretty explicitly incompatible with the MIT License which is why it was banned. Again, I'm not a lawyer.\n\n\n",
    "948014": "The deadline for the competition is comming but I am still confused. 😨 \n\nIt doesn't seem to be just a problem with yolov5.",
    "947219": ""
  }
}