{
  "id": 108491,
  "title": "Don't know how to start!!! Please help",
  "url": "/competitions/understanding_cloud_organization/discussion/108491",
  "author_name": "",
  "post_date": "2019-09-12T05:26:02.718938600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n\n<p>I am a beginner to computer vision competitions, (my prior experience is just running a unet for semantic segmentation on Carvana). Their is so much to do .. </p>\n\n<p>1) Models-&gt; Unets, MaskRCNN, FastRCNN, RCNN, \n2) Losses-&gt; Jakard, dice, focal \n3) Ensembling of models \n4) Making additional labels, pseudolabelling etc.</p>\n\n<p><strong>I feel confused what to do after reading research papers (cannot understand codes)</strong>\nThis is making me confused as how to actually start!!!!Would be grateful for your comments as to how i can go towards this in a systematic manner ( old competitions, etc).</p>\n\n<p>Thanks </p>",
  "messages": [
    {
      "id": "624457",
      "postDate": "09/12/2019 05:26:02",
      "content": "<p>Hello everyone,</p>\n\n<p>I am a beginner to computer vision competitions, (my prior experience is just running a unet for semantic segmentation on Carvana). Their is so much to do .. </p>\n\n<p>1) Models-&gt; Unets, MaskRCNN, FastRCNN, RCNN, \n2) Losses-&gt; Jakard, dice, focal \n3) Ensembling of models \n4) Making additional labels, pseudolabelling etc.</p>\n\n<p><strong>I feel confused what to do after reading research papers (cannot understand codes)</strong>\nThis is making me confused as how to actually start!!!!Would be grateful for your comments as to how i can go towards this in a systematic manner ( old competitions, etc).</p>\n\n<p>Thanks </p>",
      "rawMarkdown": "Hello everyone,\n\nI am a beginner to computer vision competitions, (my prior experience is just running a unet for semantic segmentation on Carvana). Their is so much to do .. \n\n1) Models-&gt; Unets, MaskRCNN, FastRCNN, RCNN, \n2) Losses-&gt; Jakard, dice, focal \n3) Ensembling of models \n4) Making additional labels, pseudolabelling etc.\n\n**I feel confused what to do after reading research papers (cannot understand codes)**\nThis is making me confused as how to actually start!!!!Would be grateful for your comments as to how i can go towards this in a systematic manner ( old competitions, etc).\n\n\nThanks",
      "votes": null
    },
    {
      "id": "624583",
      "postDate": "09/12/2019 07:34:32",
      "content": "<p>I think you need first to understand your points 1) and 2) ... \n( points 3) and 4) can be done later without affecting the real process)</p>\n\n<p>A. To understand 1), perhaps just choose one model first (e.g. Unet) and search for nice blogs (many) to grasp basic understanding of the model\n[Also search for nice blogs for 2).] </p>\n\n<p>B. After that come back to the public kernels to understand real implementation from our friends.</p>\n\n<p>C. Lastly, you can read the original papers for more matured understanding. </p>\n\n<p>We can loop A-B-C and we will understand more and more. Not more that 3-5 epochs, you should feel confidence in yourself. Actually I did these same process last week :)</p>",
      "rawMarkdown": "I think you need first to understand your points 1) and 2) ... \n( points 3) and 4) can be done later without affecting the real process)\n\nA. To understand 1), perhaps just choose one model first (e.g. Unet) and search for nice blogs (many) to grasp basic understanding of the model\n[Also search for nice blogs for 2).] \n\nB. After that come back to the public kernels to understand real implementation from our friends.\n\nC. Lastly, you can read the original papers for more matured understanding. \n\nWe can loop A-B-C and we will understand more and more. Not more that 3-5 epochs, you should feel confidence in yourself. Actually I did these same process last week :)",
      "votes": null
    },
    {
      "id": "624676",
      "postDate": "09/12/2019 09:09:43",
      "content": "<p>Thanks a lot :-), Could you point me to some particular resources for this competition , Instance segmentation for eg.</p>",
      "rawMarkdown": "Thanks a lot :-), Could you point me to some particular resources for this competition , Instance segmentation for eg.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 624583,
      "author_name": "ratthachat",
      "author_url": "",
      "post_date": "09/12/2019 07:34:32",
      "content": "<p>I think you need first to understand your points 1) and 2) ... \n( points 3) and 4) can be done later without affecting the real process)</p>\n\n<p>A. To understand 1), perhaps just choose one model first (e.g. Unet) and search for nice blogs (many) to grasp basic understanding of the model\n[Also search for nice blogs for 2).] </p>\n\n<p>B. After that come back to the public kernels to understand real implementation from our friends.</p>\n\n<p>C. Lastly, you can read the original papers for more matured understanding. </p>\n\n<p>We can loop A-B-C and we will understand more and more. Not more that 3-5 epochs, you should feel confidence in yourself. Actually I did these same process last week :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 624676,
      "author_name": "bashturtle",
      "author_url": "",
      "post_date": "09/12/2019 09:09:43",
      "content": "<p>Thanks a lot :-), Could you point me to some particular resources for this competition , Instance segmentation for eg.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "624457": "Hello everyone,\n\nI am a beginner to computer vision competitions, (my prior experience is just running a unet for semantic segmentation on Carvana). Their is so much to do .. \n\n1) Models-&gt; Unets, MaskRCNN, FastRCNN, RCNN, \n2) Losses-&gt; Jakard, dice, focal \n3) Ensembling of models \n4) Making additional labels, pseudolabelling etc.\n\n**I feel confused what to do after reading research papers (cannot understand codes)**\nThis is making me confused as how to actually start!!!!Would be grateful for your comments as to how i can go towards this in a systematic manner ( old competitions, etc).\n\n\nThanks",
    "624583": "I think you need first to understand your points 1) and 2) ... \n( points 3) and 4) can be done later without affecting the real process)\n\nA. To understand 1), perhaps just choose one model first (e.g. Unet) and search for nice blogs (many) to grasp basic understanding of the model\n[Also search for nice blogs for 2).] \n\nB. After that come back to the public kernels to understand real implementation from our friends.\n\nC. Lastly, you can read the original papers for more matured understanding. \n\nWe can loop A-B-C and we will understand more and more. Not more that 3-5 epochs, you should feel confidence in yourself. Actually I did these same process last week :)",
    "624676": "Thanks a lot :-), Could you point me to some particular resources for this competition , Instance segmentation for eg."
  },
  "source": "meta"
}