{
  "id": 313901,
  "title": "Can someone explain on KS's TF deep learning model?",
  "url": "/competitions/happy-whale-and-dolphin/discussion/313901",
  "author_name": "",
  "post_date": "2022-03-19T15:22:35.884026300Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello! I'm learning on KS's baseline code : <a href=\"https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu\" target=\"_blank\">https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu</a><br>\nand I cannot understand it's deep learning model.</p>\n<p>Question : Why he(or her?) is set output shape is 2048 on layer of (efficientnet-b5, global_average_pooling2d, dropout)?</p>\n<table>\n<thead>\n<tr>\n<th>Layer (type)</th>\n<th>Output Shape</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>inp1 (InputLayer)</td>\n<td>[(None, 128, 128, 3) 0</td>\n</tr>\n<tr>\n<td>efficientnet-b5 (Functional)</td>\n<td>(None, None, None, 2</td>\n</tr>\n<tr>\n<td>global_average_pooling2d</td>\n<td>(None, 2048)</td>\n</tr>\n<tr>\n<td>dropout (Dropout)</td>\n<td>(None, 2048)</td>\n</tr>\n<tr>\n<td>dense (Dense)</td>\n<td>(None, 512)</td>\n</tr>\n<tr>\n<td>inp2 (InputLayer)</td>\n<td>[(None,)]</td>\n</tr>\n<tr>\n<td>head/arcface (ArcMarginProduct)</td>\n<td>(None, 15587)</td>\n</tr>\n<tr>\n<td>softmax (Softmax)</td>\n<td>(None, 15587)</td>\n</tr>\n</tbody>\n</table>\n<p>I can't find ANY CONTEXT of this. I've found on 'efficientnot thesis', googling, and other's comment on notebook code. as I think comment on notebook code, and I found there is no reply on notebook code after a few day, so I ask on discussion.</p>\n<p>Thanks in advance for reading, and have a nice day!!</p>\n<p>P.S. Pray for Ukraine.</p>",
  "messages": [
    {
      "id": "1729054",
      "postDate": "03/19/2022 15:22:35",
      "content": "<p>Hello! I'm learning on KS's baseline code : <a href=\"https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu\" target=\"_blank\">https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu</a><br>\nand I cannot understand it's deep learning model.</p>\n<p>Question : Why he(or her?) is set output shape is 2048 on layer of (efficientnet-b5, global_average_pooling2d, dropout)?</p>\n<table>\n<thead>\n<tr>\n<th>Layer (type)</th>\n<th>Output Shape</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>inp1 (InputLayer)</td>\n<td>[(None, 128, 128, 3) 0</td>\n</tr>\n<tr>\n<td>efficientnet-b5 (Functional)</td>\n<td>(None, None, None, 2</td>\n</tr>\n<tr>\n<td>global_average_pooling2d</td>\n<td>(None, 2048)</td>\n</tr>\n<tr>\n<td>dropout (Dropout)</td>\n<td>(None, 2048)</td>\n</tr>\n<tr>\n<td>dense (Dense)</td>\n<td>(None, 512)</td>\n</tr>\n<tr>\n<td>inp2 (InputLayer)</td>\n<td>[(None,)]</td>\n</tr>\n<tr>\n<td>head/arcface (ArcMarginProduct)</td>\n<td>(None, 15587)</td>\n</tr>\n<tr>\n<td>softmax (Softmax)</td>\n<td>(None, 15587)</td>\n</tr>\n</tbody>\n</table>\n<p>I can't find ANY CONTEXT of this. I've found on 'efficientnot thesis', googling, and other's comment on notebook code. as I think comment on notebook code, and I found there is no reply on notebook code after a few day, so I ask on discussion.</p>\n<p>Thanks in advance for reading, and have a nice day!!</p>\n<p>P.S. Pray for Ukraine.</p>",
      "rawMarkdown": "Hello! I'm learning on KS's baseline code : https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu\nand I cannot understand it's deep learning model.\n\nQuestion : Why he(or her?) is set output shape is 2048 on layer of (efficientnet-b5, global_average_pooling2d, dropout)?\n\n|Layer (type)   |Output Shape        |\n| --- | --- |\n|  inp1 (InputLayer)|  [(None, 128, 128, 3) 0   |\n|efficientnet-b5 (Functional)|(None, None, None, 2 |\n|global_average_pooling2d| (None, 2048)|\n|dropout (Dropout)|(None, 2048) |\n|dense (Dense) |(None, 512)|\n|inp2 (InputLayer)|[(None,)]|\n|head/arcface (ArcMarginProduct)|(None, 15587) |\n|softmax (Softmax)| (None, 15587)|\n\nI can't find ANY CONTEXT of this. I've found on 'efficientnot thesis', googling, and other's comment on notebook code. as I think comment on notebook code, and I found there is no reply on notebook code after a few day, so I ask on discussion.\n\nThanks in advance for reading, and have a nice day!!\n\nP.S. Pray for Ukraine.",
      "votes": null
    },
    {
      "id": "1729211",
      "postDate": "03/19/2022 19:12:07",
      "content": "<p><a href=\"https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142\" target=\"_blank\">https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142</a></p>\n<p>you will see<br>\nB0:1280<br>\n…<br>\nB5:2048</p>",
      "rawMarkdown": "https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142\n\nyou will see\nB0:1280\n...\nB5:2048",
      "votes": null
    },
    {
      "id": "1729439",
      "postDate": "03/20/2022 04:53:48",
      "content": "<p>Why am I didn't get idea about see summary about Efficient-net model? Now I'm get a idea, and your comment gets me a lot of help!<br>\nthank you very much! have a nice day!</p>",
      "rawMarkdown": "Why am I didn't get idea about see summary about Efficient-net model? Now I'm get a idea, and your comment gets me a lot of help!\nthank you very much! have a nice day!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1729211,
      "author_name": "atom1231",
      "author_url": "",
      "post_date": "03/19/2022 19:12:07",
      "content": "<p><a href=\"https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142\" target=\"_blank\">https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142</a></p>\n<p>you will see<br>\nB0:1280<br>\n…<br>\nB5:2048</p>",
      "votes": null,
      "replies": [
        {
          "id": 1729439,
          "author_name": "titanicmaster",
          "author_url": "",
          "post_date": "03/20/2022 04:53:48",
          "content": "<p>Why am I didn't get idea about see summary about Efficient-net model? Now I'm get a idea, and your comment gets me a lot of help!<br>\nthank you very much! have a nice day!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1729054": "Hello! I'm learning on KS's baseline code : https://www.kaggle.com/code/ks2019/happywhale-arcface-baseline-tpu\nand I cannot understand it's deep learning model.\n\nQuestion : Why he(or her?) is set output shape is 2048 on layer of (efficientnet-b5, global_average_pooling2d, dropout)?\n\n|Layer (type)   |Output Shape        |\n| --- | --- |\n|  inp1 (InputLayer)|  [(None, 128, 128, 3) 0   |\n|efficientnet-b5 (Functional)|(None, None, None, 2 |\n|global_average_pooling2d| (None, 2048)|\n|dropout (Dropout)|(None, 2048) |\n|dense (Dense) |(None, 512)|\n|inp2 (InputLayer)|[(None,)]|\n|head/arcface (ArcMarginProduct)|(None, 15587) |\n|softmax (Softmax)| (None, 15587)|\n\nI can't find ANY CONTEXT of this. I've found on 'efficientnot thesis', googling, and other's comment on notebook code. as I think comment on notebook code, and I found there is no reply on notebook code after a few day, so I ask on discussion.\n\nThanks in advance for reading, and have a nice day!!\n\nP.S. Pray for Ukraine.",
    "1729211": "https://towardsdatascience.com/complete-architectural-details-of-all-efficientnet-models-5fd5b736142\n\nyou will see\nB0:1280\n...\nB5:2048",
    "1729439": "Why am I didn't get idea about see summary about Efficient-net model? Now I'm get a idea, and your comment gets me a lot of help!\nthank you very much! have a nice day!"
  },
  "source": "meta"
}