{
  "id": 134633,
  "title": "who trained GhostNet? 5 models inference ~6min",
  "url": "/competitions/bengaliai-cv19/discussion/134633",
  "author_name": "Alex Shonenkov",
  "post_date": "2020-03-09T11:15:28.002000",
  "votes": 8,
  "comment_count": 8,
  "views": 0,
  "content": "<p>ghostnet pytorch implemetation:\n<a href=\"https://github.com/iamhankai/ghostnet.pytorch\">https://github.com/iamhankai/ghostnet.pytorch</a></p>\n\n<p>my inference:\n<a href=\"https://www.kaggle.com/shonenkov/ghostnet-super-easy-inference\">https://www.kaggle.com/shonenkov/ghostnet-super-easy-inference</a></p>\n\n<p>Public LB: 0.9702</p>\n\n<p>who trained GhostNet?</p>",
  "messages": [
    {
      "id": 767245,
      "postDate": "2020-03-09T11:15:28.003Z",
      "content": "<p>ghostnet pytorch implemetation:\n<a href=\"https://github.com/iamhankai/ghostnet.pytorch\">https://github.com/iamhankai/ghostnet.pytorch</a></p>\n\n<p>my inference:\n<a href=\"https://www.kaggle.com/shonenkov/ghostnet-super-easy-inference\">https://www.kaggle.com/shonenkov/ghostnet-super-easy-inference</a></p>\n\n<p>Public LB: 0.9702</p>\n\n<p>who trained GhostNet?</p>",
      "rawMarkdown": "ghostnet pytorch implemetation:\nhttps://github.com/iamhankai/ghostnet.pytorch\n\nmy inference:\nhttps://www.kaggle.com/shonenkov/ghostnet-super-easy-inference\n\nPublic LB: 0.9702\n\nwho trained GhostNet?",
      "votes": 8
    },
    {
      "id": 767763,
      "postDate": "2020-03-10T04:27:25.520Z",
      "content": "<p>I got 0.9741 with ghostnet</p>",
      "rawMarkdown": "I got 0.9741 with ghostnet",
      "votes": 1,
      "replies": [
        {
          "id": 767872,
          "postDate": "2020-03-10T07:35:54.433Z",
          "content": "<p>👍  nice score</p>",
          "rawMarkdown": "👍  nice score"
        },
        {
          "id": 933613,
          "postDate": "2020-07-17T20:22:39.380Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 767365,
      "postDate": "2020-03-09T14:34:03.857Z",
      "content": "<p>is the pretrained model available?</p>",
      "rawMarkdown": "is the pretrained model available?\n",
      "votes": 1,
      "replies": [
        {
          "id": 767470,
          "postDate": "2020-03-09T17:16:04.113Z",
          "content": "<p>Maybe, but I didn't use ghostnet pretrained model</p>",
          "rawMarkdown": "Maybe, but I didn't use ghostnet pretrained model"
        },
        {
          "id": 767527,
          "postDate": "2020-03-09T18:48:55.847Z",
          "content": "<p>There is a TensorFlow version pretrained model.\n<a href=\"https://github.com/huawei-noah/ghostnet\">https://github.com/huawei-noah/ghostnet</a></p>\n\n<p>Requirements\nThe code was verified on Python3.6, TensorFlow-1.13.1, Tensorpack-0.9.7. Not sure on other version.</p>\n\n<p>Usage\nRun python main.py --eval --data_dir=/path/to/imagenet/dir/ --load=./models/ghostnet_checkpoint to evaluate on val set.</p>\n\n<p>You'll get the accuracy: top-1 error=0.26066, top-5 error=0.08614 with only 141M Flops (or say MAdds).</p>",
          "rawMarkdown": "There is a TensorFlow version pretrained model.\nhttps://github.com/huawei-noah/ghostnet\n\nRequirements\nThe code was verified on Python3.6, TensorFlow-1.13.1, Tensorpack-0.9.7. Not sure on other version.\n\nUsage\nRun python main.py --eval --data_dir=/path/to/imagenet/dir/ --load=./models/ghostnet_checkpoint to evaluate on val set.\n\nYou'll get the accuracy: top-1 error=0.26066, top-5 error=0.08614 with only 141M Flops (or say MAdds)."
        },
        {
          "id": 767651,
          "postDate": "2020-03-10T00:40:15.067Z",
          "content": "<p>i'm training ghostnet from scratch but the performance is quite low for a model that can achieve same performance with B0 on ImageNet, maybe it's because we dont have pretrained weights for pytorch</p>",
          "rawMarkdown": "i'm training ghostnet from scratch but the performance is quite low for a model that can achieve same performance with B0 on ImageNet, maybe it's because we dont have pretrained weights for pytorch"
        }
      ]
    },
    {
      "id": 933610,
      "postDate": "2020-07-17T20:21:41.650Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 767763,
      "author_name": "Jaebb",
      "author_url": "",
      "post_date": "2020-03-10T04:27:25.520000",
      "content": "<p>I got 0.9741 with ghostnet</p>",
      "votes": 1,
      "replies": [
        {
          "id": 767872,
          "author_name": "Alex Shonenkov",
          "author_url": "",
          "post_date": "2020-03-10T07:35:54.433000",
          "content": "<p>👍  nice score</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 933613,
          "author_name": "",
          "author_url": "",
          "post_date": "2020-07-17T20:22:39.380000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 767365,
      "author_name": "DatNT",
      "author_url": "",
      "post_date": "2020-03-09T14:34:03.857000",
      "content": "<p>is the pretrained model available?</p>",
      "votes": 1,
      "replies": [
        {
          "id": 767470,
          "author_name": "Alex Shonenkov",
          "author_url": "",
          "post_date": "2020-03-09T17:16:04.113000",
          "content": "<p>Maybe, but I didn't use ghostnet pretrained model</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 767527,
          "author_name": "Helen",
          "author_url": "",
          "post_date": "2020-03-09T18:48:55.847000",
          "content": "<p>There is a TensorFlow version pretrained model.\n<a href=\"https://github.com/huawei-noah/ghostnet\">https://github.com/huawei-noah/ghostnet</a></p>\n\n<p>Requirements\nThe code was verified on Python3.6, TensorFlow-1.13.1, Tensorpack-0.9.7. Not sure on other version.</p>\n\n<p>Usage\nRun python main.py --eval --data_dir=/path/to/imagenet/dir/ --load=./models/ghostnet_checkpoint to evaluate on val set.</p>\n\n<p>You'll get the accuracy: top-1 error=0.26066, top-5 error=0.08614 with only 141M Flops (or say MAdds).</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 767651,
          "author_name": "DatNT",
          "author_url": "",
          "post_date": "2020-03-10T00:40:15.067000",
          "content": "<p>i'm training ghostnet from scratch but the performance is quite low for a model that can achieve same performance with B0 on ImageNet, maybe it's because we dont have pretrained weights for pytorch</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 933610,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-07-17T20:21:41.650000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "767245": "ghostnet pytorch implemetation:\nhttps://github.com/iamhankai/ghostnet.pytorch\n\nmy inference:\nhttps://www.kaggle.com/shonenkov/ghostnet-super-easy-inference\n\nPublic LB: 0.9702\n\nwho trained GhostNet?",
    "767763": "I got 0.9741 with ghostnet",
    "767365": "is the pretrained model available?\n",
    "933610": ""
  }
}