{
  "id": 104246,
  "title": "LB 54 with PatchGAN",
  "url": "/competitions/generative-dog-images/discussion/104246",
  "author_name": "",
  "post_date": "2019-08-15T13:34:13.646450200Z",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Noob here, don't know much about DL, here to flap arms and legs to learn swimming in the DeepOcean.</p>\n\n<p>I started with Memorization GANs and tried to learn by improving upon using @cdeotte's kernel. Then memorization GANs were banned and when @phoenix9032 uploaded his 60LB kernel and @timetraveller98 started the \"How to train your touchy GANs - Things that seem to work\" discussion, I tried to learn from them too. Looked here and there and read about a few GAN architectures notably ProGAN and how they used a PatchGAN to generate sharper images. Used that one here along with augmentation and randomly flips of labels at training time. Couldn't incorporate other tips, still learning :D</p>\n\n<p>Thanks to all!</p>\n\n<p>Here's the kernel: <a href=\"https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862\">https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862</a></p>",
  "messages": [
    {
      "id": "599951",
      "postDate": "08/15/2019 13:34:13",
      "content": "<p>Noob here, don't know much about DL, here to flap arms and legs to learn swimming in the DeepOcean.</p>\n\n<p>I started with Memorization GANs and tried to learn by improving upon using @cdeotte's kernel. Then memorization GANs were banned and when @phoenix9032 uploaded his 60LB kernel and @timetraveller98 started the \"How to train your touchy GANs - Things that seem to work\" discussion, I tried to learn from them too. Looked here and there and read about a few GAN architectures notably ProGAN and how they used a PatchGAN to generate sharper images. Used that one here along with augmentation and randomly flips of labels at training time. Couldn't incorporate other tips, still learning :D</p>\n\n<p>Thanks to all!</p>\n\n<p>Here's the kernel: <a href=\"https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862\">https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862</a></p>",
      "rawMarkdown": "Noob here, don't know much about DL, here to flap arms and legs to learn swimming in the DeepOcean.\n\nI started with Memorization GANs and tried to learn by improving upon using @cdeotte's kernel. Then memorization GANs were banned and when @phoenix9032 uploaded his 60LB kernel and @timetraveller98 started the \"How to train your touchy GANs - Things that seem to work\" discussion, I tried to learn from them too. Looked here and there and read about a few GAN architectures notably ProGAN and how they used a PatchGAN to generate sharper images. Used that one here along with augmentation and randomly flips of labels at training time. Couldn't incorporate other tips, still learning :D\n\nThanks to all!\n\nHere's the kernel: https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862",
      "votes": null
    },
    {
      "id": "599976",
      "postDate": "08/15/2019 13:51:04",
      "content": "<p>Nice work Mighty Rains. You did well improving Nirjhar's kernel from LB 60 to LB 54</p>",
      "rawMarkdown": "Nice work Mighty Rains. You did well improving Nirjhar's kernel from LB 60 to LB 54",
      "votes": null
    },
    {
      "id": "600081",
      "postDate": "08/15/2019 16:10:55",
      "content": "<p>Thanks a ton! It was your kernel that got me started here :)</p>",
      "rawMarkdown": "Thanks a ton! It was your kernel that got me started here :)",
      "votes": null
    },
    {
      "id": "600401",
      "postDate": "08/16/2019 04:21:27",
      "content": "<p>Great..... Nice Share...Thanks.. <a href=\"/mightyrains\">@mightyrains</a> </p>",
      "rawMarkdown": "Great..... Nice Share...Thanks.. @mightyrains",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 599976,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "08/15/2019 13:51:04",
      "content": "<p>Nice work Mighty Rains. You did well improving Nirjhar's kernel from LB 60 to LB 54</p>",
      "votes": null,
      "replies": [
        {
          "id": 600081,
          "author_name": "mightyrains",
          "author_url": "",
          "post_date": "08/15/2019 16:10:55",
          "content": "<p>Thanks a ton! It was your kernel that got me started here :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 600401,
      "author_name": "veeralakrishna",
      "author_url": "",
      "post_date": "08/16/2019 04:21:27",
      "content": "<p>Great..... Nice Share...Thanks.. <a href=\"/mightyrains\">@mightyrains</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "599951": "Noob here, don't know much about DL, here to flap arms and legs to learn swimming in the DeepOcean.\n\nI started with Memorization GANs and tried to learn by improving upon using @cdeotte's kernel. Then memorization GANs were banned and when @phoenix9032 uploaded his 60LB kernel and @timetraveller98 started the \"How to train your touchy GANs - Things that seem to work\" discussion, I tried to learn from them too. Looked here and there and read about a few GAN architectures notably ProGAN and how they used a PatchGAN to generate sharper images. Used that one here along with augmentation and randomly flips of labels at training time. Couldn't incorporate other tips, still learning :D\n\nThanks to all!\n\nHere's the kernel: https://www.kaggle.com/mightyrains/patchdogs-2x2?scriptVersionId=18792862",
    "599976": "Nice work Mighty Rains. You did well improving Nirjhar's kernel from LB 60 to LB 54",
    "600081": "Thanks a ton! It was your kernel that got me started here :)",
    "600401": "Great..... Nice Share...Thanks.. @mightyrains"
  },
  "source": "meta"
}