{
  "id": 103061,
  "title": "How does shape of input noise effect GAN?",
  "url": "/competitions/generative-dog-images/discussion/103061",
  "author_name": "",
  "post_date": "2019-08-06T21:04:16.372239900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Does the shape of the input noise(z) given to the generator have any effect on the quality of generated images or training process. Please can anyone explain with a reason?</p>\n\n<p>Most of the GAN architectures use (100,) as their input shapes, so initially I was using the same but when I changed it to (128,) it made some improvements to my results. </p>",
  "messages": [
    {
      "id": "593600",
      "postDate": "08/06/2019 21:04:16",
      "content": "<p>Does the shape of the input noise(z) given to the generator have any effect on the quality of generated images or training process. Please can anyone explain with a reason?</p>\n\n<p>Most of the GAN architectures use (100,) as their input shapes, so initially I was using the same but when I changed it to (128,) it made some improvements to my results. </p>",
      "rawMarkdown": "Does the shape of the input noise(z) given to the generator have any effect on the quality of generated images or training process. Please can anyone explain with a reason?\n\nMost of the GAN architectures use (100,) as their input shapes, so initially I was using the same but when I changed it to (128,) it made some improvements to my results.",
      "votes": null
    },
    {
      "id": "596948",
      "postDate": "08/11/2019 15:11:17",
      "content": "<p>There is an influence on the size of noise (vector). I have done face generation before, and the size taken is (218,178). You may refer to my GitHub,<a href=\"https://github.com/Lilhxr/NUS_GAN_Final\">https://github.com/Lilhxr/NUS_GAN_Final</a></p>",
      "rawMarkdown": "There is an influence on the size of noise (vector). I have done face generation before, and the size taken is (218,178). You may refer to my GitHub,https://github.com/Lilhxr/NUS_GAN_Final",
      "votes": null
    },
    {
      "id": "599059",
      "postDate": "08/14/2019 12:46:50",
      "content": "<p>Yes, the dimensions of input (random noise vector)  does decide the output. Since the generator is producing output from the input, the input dimensions plays a major role governing it.The larger the dimensions of input vector, more better output can be generated. Generally, for ease size of input is taken as 100, but yes we can take any size for the noise vector.</p>",
      "rawMarkdown": "Yes, the dimensions of input (random noise vector)  does decide the output. Since the generator is producing output from the input, the input dimensions plays a major role governing it.The larger the dimensions of input vector, more better output can be generated. Generally, for ease size of input is taken as 100, but yes we can take any size for the noise vector.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 596948,
      "author_name": "",
      "author_url": "",
      "post_date": "08/11/2019 15:11:17",
      "content": "<p>There is an influence on the size of noise (vector). I have done face generation before, and the size taken is (218,178). You may refer to my GitHub,<a href=\"https://github.com/Lilhxr/NUS_GAN_Final\">https://github.com/Lilhxr/NUS_GAN_Final</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 599059,
      "author_name": "samriddhiraut",
      "author_url": "",
      "post_date": "08/14/2019 12:46:50",
      "content": "<p>Yes, the dimensions of input (random noise vector)  does decide the output. Since the generator is producing output from the input, the input dimensions plays a major role governing it.The larger the dimensions of input vector, more better output can be generated. Generally, for ease size of input is taken as 100, but yes we can take any size for the noise vector.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "593600": "Does the shape of the input noise(z) given to the generator have any effect on the quality of generated images or training process. Please can anyone explain with a reason?\n\nMost of the GAN architectures use (100,) as their input shapes, so initially I was using the same but when I changed it to (128,) it made some improvements to my results.",
    "596948": "There is an influence on the size of noise (vector). I have done face generation before, and the size taken is (218,178). You may refer to my GitHub,https://github.com/Lilhxr/NUS_GAN_Final",
    "599059": "Yes, the dimensions of input (random noise vector)  does decide the output. Since the generator is producing output from the input, the input dimensions plays a major role governing it.The larger the dimensions of input vector, more better output can be generated. Generally, for ease size of input is taken as 100, but yes we can take any size for the noise vector."
  },
  "source": "meta"
}