{
  "id": 196946,
  "title": "\"cb50326950.jpg\" and \"c78b4fa3a9.jpg\" are rounded.",
  "url": "/competitions/gan-getting-started/discussion/196946",
  "author_name": "",
  "post_date": "2020-11-13T14:08:35.937218500Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4705965%2F166d84bc10194308ba4881c60199d5b7%2F(1).png?generation=1605276381833086&amp;alt=media\" alt=\"\"><br>\nleft: <code>monet_jpg/cb50326950.jpg</code>, right: <code>monet_jpg/c78b4fa3a9.jpg</code></p>\n<p>I noticed something not so good about the two images.<br>\nThey may have a negative effect on learning the GAN Discriminator.</p>",
  "messages": [
    {
      "id": "1077356",
      "postDate": "11/13/2020 14:08:35",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4705965%2F166d84bc10194308ba4881c60199d5b7%2F(1).png?generation=1605276381833086&amp;alt=media\" alt=\"\"><br>\nleft: <code>monet_jpg/cb50326950.jpg</code>, right: <code>monet_jpg/c78b4fa3a9.jpg</code></p>\n<p>I noticed something not so good about the two images.<br>\nThey may have a negative effect on learning the GAN Discriminator.</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4705965%2F166d84bc10194308ba4881c60199d5b7%2F(1).png?generation=1605276381833086&alt=media)\nleft: `monet_jpg/cb50326950.jpg`, right: `monet_jpg/c78b4fa3a9.jpg`\n\nI noticed something not so good about the two images.\nThey may have a negative effect on learning the GAN Discriminator.",
      "votes": null
    },
    {
      "id": "1077445",
      "postDate": "11/13/2020 16:13:27",
      "content": "<p>Yes, I noticed those anomalies too, but I think it shouldn't be a problem because the deeper hidden CNN layer of GAN will extract the complex style representative features (eg textures, color composition, contrast e.t.c )of the monet images.<br>\nIf you still don't want that circular border, you may apply center crop and rescale those specific images.</p>",
      "rawMarkdown": "Yes, I noticed those anomalies too, but I think it shouldn't be a problem because the deeper hidden CNN layer of GAN will extract the complex style representative features (eg textures, color composition, contrast e.t.c )of the monet images.\nIf you still don't want that circular border, you may apply center crop and rescale those specific images.",
      "votes": null
    },
    {
      "id": "1077727",
      "postDate": "11/13/2020 21:33:24",
      "content": "<p>Thanks you for answer.👍<br>\nMy anxious is the patch Discriminator on CycleGAN has sightly shallow layers than other DNNs.</p>\n<p>I think it mitigate when I will adopt \"<strong>Mixup Augmentation</strong>” in GAN training. <br>\nOR I'll simply delete two images.</p>",
      "rawMarkdown": "Thanks you for answer.👍\nMy anxious is the patch Discriminator on CycleGAN has sightly shallow layers than other DNNs.\n\nI think it mitigate when I will adopt \"**Mixup Augmentation**” in GAN training. \nOR I'll simply delete two images.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1077445,
      "author_name": "parikshitsaikia1619",
      "author_url": "",
      "post_date": "11/13/2020 16:13:27",
      "content": "<p>Yes, I noticed those anomalies too, but I think it shouldn't be a problem because the deeper hidden CNN layer of GAN will extract the complex style representative features (eg textures, color composition, contrast e.t.c )of the monet images.<br>\nIf you still don't want that circular border, you may apply center crop and rescale those specific images.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1077727,
          "author_name": "stealthflow",
          "author_url": "",
          "post_date": "11/13/2020 21:33:24",
          "content": "<p>Thanks you for answer.👍<br>\nMy anxious is the patch Discriminator on CycleGAN has sightly shallow layers than other DNNs.</p>\n<p>I think it mitigate when I will adopt \"<strong>Mixup Augmentation</strong>” in GAN training. <br>\nOR I'll simply delete two images.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1077356": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4705965%2F166d84bc10194308ba4881c60199d5b7%2F(1).png?generation=1605276381833086&alt=media)\nleft: `monet_jpg/cb50326950.jpg`, right: `monet_jpg/c78b4fa3a9.jpg`\n\nI noticed something not so good about the two images.\nThey may have a negative effect on learning the GAN Discriminator.",
    "1077445": "Yes, I noticed those anomalies too, but I think it shouldn't be a problem because the deeper hidden CNN layer of GAN will extract the complex style representative features (eg textures, color composition, contrast e.t.c )of the monet images.\nIf you still don't want that circular border, you may apply center crop and rescale those specific images.",
    "1077727": "Thanks you for answer.👍\nMy anxious is the patch Discriminator on CycleGAN has sightly shallow layers than other DNNs.\n\nI think it mitigate when I will adopt \"**Mixup Augmentation**” in GAN training. \nOR I'll simply delete two images."
  },
  "source": "meta"
}