{
  "id": 58490,
  "title": "ConvertVideoToCsv",
  "url": "/competitions/cvpr-2018-autonomous-driving/discussion/58490",
  "author_name": "",
  "post_date": "2018-06-08T20:27:19.738250300Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I am now trying to use ConvertVideotoCsv.py file in order to convert my binary masks into a submission file. Previously I used my own script for this, so now I want to check if it is really correct. However, I have some questions about this.\n1. Do I need to have for each object instance a separate image for a binary mask?\n2. How should I name this images of binary masks? \nDo I use for this test_video_list_and_name_mapping.zip files?\nCould you please provide some examples and/or any additional script (in python) that you used to accomplish this?\nThank you in advance.</p>",
  "messages": [
    {
      "id": "340296",
      "postDate": "06/08/2018 20:27:19",
      "content": "<p>I am now trying to use ConvertVideotoCsv.py file in order to convert my binary masks into a submission file. Previously I used my own script for this, so now I want to check if it is really correct. However, I have some questions about this.\n1. Do I need to have for each object instance a separate image for a binary mask?\n2. How should I name this images of binary masks? \nDo I use for this test_video_list_and_name_mapping.zip files?\nCould you please provide some examples and/or any additional script (in python) that you used to accomplish this?\nThank you in advance.</p>",
      "rawMarkdown": "I am now trying to use ConvertVideotoCsv.py file in order to convert my binary masks into a submission file. Previously I used my own script for this, so now I want to check if it is really correct. However, I have some questions about this.\n1. Do I need to have for each object instance a separate image for a binary mask?\n2. How should I name this images of binary masks? \nDo I use for this test_video_list_and_name_mapping.zip files?\nCould you please provide some examples and/or any additional script (in python) that you used to accomplish this?\nThank you in advance.",
      "votes": null
    },
    {
      "id": "340361",
      "postDate": "06/09/2018 02:30:11",
      "content": "<p>All of your questions are basically answered in the comments of the ConvertVideotoCsv.py file.</p>\n\n<ol>\n<li>Yes, if you want to use the provided script (ConvertVideotoCsv.py)</li>\n<li>I recommend something like \"Image name\"_\"instance id\".png</li>\n<li>I used those as starting points to get the names of the files</li>\n<li>I can share my script I used for this process, it is just an adapted version of ConvertVideotoCsv.py which takes advantage of multiple CPUs. I no longer use it though because it is too time consuming to write all instances maskes to the disk only to retrieve them later. I recommend converting the network output directly into the required CSV submission format.</li>\n</ol>",
      "rawMarkdown": "All of your questions are basically answered in the comments of the ConvertVideotoCsv.py file.\n\n1. Yes, if you want to use the provided script (ConvertVideotoCsv.py)\n2. I recommend something like \"Image name\"_\"instance id\".png\n3.  I used those as starting points to get the names of the files\n4. I can share my script I used for this process, it is just an adapted version of ConvertVideotoCsv.py which takes advantage of multiple CPUs. I no longer use it though because it is too time consuming to write all instances maskes to the disk only to retrieve them later. I recommend converting the network output directly into the required CSV submission format.",
      "votes": null
    },
    {
      "id": "340482",
      "postDate": "06/09/2018 09:41:24",
      "content": "<p>Thank you very much on your answer. Would you mind sharing the script that you use to directly convert your network output to the CSV submission format (without using ConvertVideotoCsv.py)? Which is the format of your network output? I am using facebook detectron software (MaskRCNN). I know that this question comes a bit late, but I would like to check if I am really getting the right format. My results are around 0.05, which is a bit too low.</p>",
      "rawMarkdown": "Thank you very much on your answer. Would you mind sharing the script that you use to directly convert your network output to the CSV submission format (without using ConvertVideotoCsv.py)? Which is the format of your network output? I am using facebook detectron software (MaskRCNN). I know that this question comes a bit late, but I would like to check if I am really getting the right format. My results are around 0.05, which is a bit too low.",
      "votes": null
    },
    {
      "id": "340602",
      "postDate": "06/09/2018 18:29:33",
      "content": "<p>I use the one provided in this <a href=\"https://www.kaggle.com/c/cvpr-2018-autonomous-driving/discussion/57706\">discussion</a>. It is based on output from MaskRCNN, which I use as well. If you get any score over 0.0, then your script is probably correct, however if your model is small then you won't get much more than 0.05.</p>",
      "rawMarkdown": "I use the one provided in this [discussion][1]. It is based on output from MaskRCNN, which I use as well. If you get any score over 0.0, then your script is probably correct, however if your model is small then you won't get much more than 0.05.\n\n\n  [1]: https://www.kaggle.com/c/cvpr-2018-autonomous-driving/discussion/57706",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 340361,
      "author_name": "michaelheinzer",
      "author_url": "",
      "post_date": "06/09/2018 02:30:11",
      "content": "<p>All of your questions are basically answered in the comments of the ConvertVideotoCsv.py file.</p>\n\n<ol>\n<li>Yes, if you want to use the provided script (ConvertVideotoCsv.py)</li>\n<li>I recommend something like \"Image name\"_\"instance id\".png</li>\n<li>I used those as starting points to get the names of the files</li>\n<li>I can share my script I used for this process, it is just an adapted version of ConvertVideotoCsv.py which takes advantage of multiple CPUs. I no longer use it though because it is too time consuming to write all instances maskes to the disk only to retrieve them later. I recommend converting the network output directly into the required CSV submission format.</li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 340482,
          "author_name": "natasa1",
          "author_url": "",
          "post_date": "06/09/2018 09:41:24",
          "content": "<p>Thank you very much on your answer. Would you mind sharing the script that you use to directly convert your network output to the CSV submission format (without using ConvertVideotoCsv.py)? Which is the format of your network output? I am using facebook detectron software (MaskRCNN). I know that this question comes a bit late, but I would like to check if I am really getting the right format. My results are around 0.05, which is a bit too low.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 340602,
          "author_name": "michaelheinzer",
          "author_url": "",
          "post_date": "06/09/2018 18:29:33",
          "content": "<p>I use the one provided in this <a href=\"https://www.kaggle.com/c/cvpr-2018-autonomous-driving/discussion/57706\">discussion</a>. It is based on output from MaskRCNN, which I use as well. If you get any score over 0.0, then your script is probably correct, however if your model is small then you won't get much more than 0.05.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "340296": "I am now trying to use ConvertVideotoCsv.py file in order to convert my binary masks into a submission file. Previously I used my own script for this, so now I want to check if it is really correct. However, I have some questions about this.\n1. Do I need to have for each object instance a separate image for a binary mask?\n2. How should I name this images of binary masks? \nDo I use for this test_video_list_and_name_mapping.zip files?\nCould you please provide some examples and/or any additional script (in python) that you used to accomplish this?\nThank you in advance.",
    "340361": "All of your questions are basically answered in the comments of the ConvertVideotoCsv.py file.\n\n1. Yes, if you want to use the provided script (ConvertVideotoCsv.py)\n2. I recommend something like \"Image name\"_\"instance id\".png\n3.  I used those as starting points to get the names of the files\n4. I can share my script I used for this process, it is just an adapted version of ConvertVideotoCsv.py which takes advantage of multiple CPUs. I no longer use it though because it is too time consuming to write all instances maskes to the disk only to retrieve them later. I recommend converting the network output directly into the required CSV submission format.",
    "340482": "Thank you very much on your answer. Would you mind sharing the script that you use to directly convert your network output to the CSV submission format (without using ConvertVideotoCsv.py)? Which is the format of your network output? I am using facebook detectron software (MaskRCNN). I know that this question comes a bit late, but I would like to check if I am really getting the right format. My results are around 0.05, which is a bit too low.",
    "340602": "I use the one provided in this [discussion][1]. It is based on output from MaskRCNN, which I use as well. If you get any score over 0.0, then your script is probably correct, however if your model is small then you won't get much more than 0.05.\n\n\n  [1]: https://www.kaggle.com/c/cvpr-2018-autonomous-driving/discussion/57706"
  },
  "source": "meta"
}