{
  "id": 89744,
  "title": "Base model select",
  "url": "/competitions/imet-2019-fgvc6/discussion/89744",
  "author_name": "",
  "post_date": "2019-04-17T06:19:19.829504600Z",
  "votes": 24,
  "comment_count": 13,
  "views": 0,
  "content": "<p>When selecting base model,  I think you can check this table.\n| Model        | Acc@1           | Acc@5  |Time* |\n| ------------- |:-------------:| -----:| -----:|\n| vgg16  | 70.79 | 89.74 |24.95|\n|vgg19 |    70.89| 89.69|   24.95|\n|resnet18|68.24|88.49|  16.07 |\n|resnet34|72.17|90.74|  17.37|\n|resnet50|74.81|92.38|  22.62|\n|resnet101|76.58|93.10|33.03|\n|resnet152|76.66|93.08|42.37|\n|resnet50v2|69.73|89.31|19.56|\n|resnet101v2|71.93|90.41|28.80|\n|resnet152v2|72.29|90.61|41.09|\n|resnext50|77.36|93.48|37.57|\n|resnext101|78.48|94.00|60.07|\n|densenet121|74.67|92.04|27.66|\n|densenet169|75.85|92.93|33.71|\n|densenet201    |77.13|93.43|42.40|\n|inceptionv3|77.55|93.48|38.94|\n|xception|78.87|94.20|42.18|\n|inceptionresnetv2|80.03|94.89|54.77|\n|seresnet18|69.41|88.84|20.19|\n|seresnet34|    72.60|90.91|22.20|\n|seresnet50|    76.44|93.02|23.64|\n|seresnet101|   77.92|94.00|32.55|\n|seresnet152|   78.34|94.08|47.88   |\n|seresnext50|   78.74|94.30|38.29|\n|seresnext101|79.88|94.87|62.80|\n|senet154|81.06|95.24|137.36|\n|nasnetlarge|   82.12|95.72|116.53|\n|nasnetmobile|74.04|91.54|27.73|\n|mobilenet|70.36|89.39|15.50|\n|mobilenetv2|71.63|90.35|18.31|</p>",
  "messages": [
    {
      "id": "518380",
      "postDate": "04/17/2019 06:19:19",
      "content": "<p>When selecting base model,  I think you can check this table.\n| Model        | Acc@1           | Acc@5  |Time* |\n| ------------- |:-------------:| -----:| -----:|\n| vgg16  | 70.79 | 89.74 |24.95|\n|vgg19 |    70.89| 89.69|   24.95|\n|resnet18|68.24|88.49|  16.07 |\n|resnet34|72.17|90.74|  17.37|\n|resnet50|74.81|92.38|  22.62|\n|resnet101|76.58|93.10|33.03|\n|resnet152|76.66|93.08|42.37|\n|resnet50v2|69.73|89.31|19.56|\n|resnet101v2|71.93|90.41|28.80|\n|resnet152v2|72.29|90.61|41.09|\n|resnext50|77.36|93.48|37.57|\n|resnext101|78.48|94.00|60.07|\n|densenet121|74.67|92.04|27.66|\n|densenet169|75.85|92.93|33.71|\n|densenet201    |77.13|93.43|42.40|\n|inceptionv3|77.55|93.48|38.94|\n|xception|78.87|94.20|42.18|\n|inceptionresnetv2|80.03|94.89|54.77|\n|seresnet18|69.41|88.84|20.19|\n|seresnet34|    72.60|90.91|22.20|\n|seresnet50|    76.44|93.02|23.64|\n|seresnet101|   77.92|94.00|32.55|\n|seresnet152|   78.34|94.08|47.88   |\n|seresnext50|   78.74|94.30|38.29|\n|seresnext101|79.88|94.87|62.80|\n|senet154|81.06|95.24|137.36|\n|nasnetlarge|   82.12|95.72|116.53|\n|nasnetmobile|74.04|91.54|27.73|\n|mobilenet|70.36|89.39|15.50|\n|mobilenetv2|71.63|90.35|18.31|</p>",
      "rawMarkdown": "When selecting base model,  I think you can check this table.\n| Model        | Acc@1           | Acc@5  |Time* |\n| ------------- |:-------------:| -----:| -----:|\n| vgg16  | 70.79 | 89.74 |24.95|\n|vgg19 |\t70.89| 89.69|\t24.95|\n|resnet18|68.24|88.49|\t16.07 |\n|resnet34|72.17|90.74|\t17.37|\n|resnet50|74.81|92.38|\t22.62|\n|resnet101|76.58|93.10|33.03|\n|resnet152|76.66|93.08|42.37|\n|resnet50v2|69.73|89.31|19.56|\n|resnet101v2|71.93|90.41|28.80|\n|resnet152v2|72.29|90.61|41.09|\n|resnext50|77.36|93.48|37.57|\n|resnext101|78.48|94.00|60.07|\n|densenet121|74.67|92.04|27.66|\n|densenet169|75.85|92.93|33.71|\n|densenet201\t|77.13|93.43|42.40|\n|inceptionv3|77.55|93.48|38.94|\n|xception|78.87|94.20|42.18|\n|inceptionresnetv2|80.03|94.89|54.77|\n|seresnet18|69.41|88.84|20.19|\n|seresnet34|\t72.60|90.91|22.20|\n|seresnet50|\t76.44|93.02|23.64|\n|seresnet101|\t77.92|94.00|32.55|\n|seresnet152|\t78.34|94.08|47.88\t|\n|seresnext50|\t78.74|94.30|38.29|\n|seresnext101|79.88|94.87|62.80|\n|senet154|81.06|95.24|137.36|\n|nasnetlarge|\t82.12|95.72|116.53|\n|nasnetmobile|74.04|91.54|27.73|\n|mobilenet|70.36|89.39|15.50|\n|mobilenetv2|71.63|90.35|18.31|",
      "votes": null
    },
    {
      "id": "518406",
      "postDate": "04/17/2019 07:02:30",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    },
    {
      "id": "518420",
      "postDate": "04/17/2019 07:30:24",
      "content": "<p>Very welcome! I have revised this discussion into table format, which would give you better reading experience!</p>",
      "rawMarkdown": "Very welcome! I have revised this discussion into table format, which would give you better reading experience!",
      "votes": null
    },
    {
      "id": "518543",
      "postDate": "04/17/2019 12:44:29",
      "content": "<p>Thanks for sharing！If you use pytorch, you can check this <a href=\"https://github.com/Cadene/pretrained-models.pytorch#accuracy-on-validation-set-single-model\">list</a> in <a href=\"https://github.com/Cadene/pretrained-models.pytorch\">pretrainedmodels</a>.</p>",
      "rawMarkdown": "Thanks for sharing！If you use pytorch, you can check this [list](https://github.com/Cadene/pretrained-models.pytorch#accuracy-on-validation-set-single-model) in [pretrainedmodels](https://github.com/Cadene/pretrained-models.pytorch).",
      "votes": null
    },
    {
      "id": "518590",
      "postDate": "04/17/2019 13:49:01",
      "content": "<p>And this kaggle dataset. <a href=\"https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo\">https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo</a></p>",
      "rawMarkdown": "And this kaggle dataset. https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo",
      "votes": null
    },
    {
      "id": "518815",
      "postDate": "04/17/2019 21:33:47",
      "content": "<p>I was just beginning to study machine learning, \nthis list of networks is very useful information for me.</p>",
      "rawMarkdown": "I was just beginning to study machine learning, \nthis list of networks is very useful information for me.",
      "votes": null
    },
    {
      "id": "520236",
      "postDate": "04/20/2019 13:58:44",
      "content": "<p>Thanks, for sharing, also for anyone using Keras checkout this link: <a href=\"https://keras.io/applications/\">https://keras.io/applications/</a></p>",
      "rawMarkdown": "Thanks, for sharing, also for anyone using Keras checkout this link: https://keras.io/applications/",
      "votes": null
    },
    {
      "id": "520243",
      "postDate": "04/20/2019 14:19:04",
      "content": "<p>Could you explain the Acc@1 vs Acc@5 columns? </p>",
      "rawMarkdown": "Could you explain the Acc@1 vs Acc@5 columns?",
      "votes": null
    },
    {
      "id": "520608",
      "postDate": "04/21/2019 12:06:00",
      "content": "<p><a href=\"https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5\">top-1 &amp; top-5 accuracy</a>These means top-1 accuracy and top-5 accuracy separately, I think this website could be helpful</p>",
      "rawMarkdown": "[top-1 &amp; top-5 accuracy](https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5)These means top-1 accuracy and top-5 accuracy separately, I think this website could be helpful",
      "votes": null
    },
    {
      "id": "520714",
      "postDate": "04/21/2019 16:14:42",
      "content": "<p>You can find more pretrained and not pretrained models at this excellent repo <a href=\"https://github.com/osmr/imgclsmob\">https://github.com/osmr/imgclsmob</a> </p>\n\n<p>I use it for pytorch, but it has models for Keras and chainer as well, though I have not tried them. </p>",
      "rawMarkdown": "You can find more pretrained and not pretrained models at this excellent repo https://github.com/osmr/imgclsmob \n\nI use it for pytorch, but it has models for Keras and chainer as well, though I have not tried them.",
      "votes": null
    },
    {
      "id": "520722",
      "postDate": "04/21/2019 16:35:26",
      "content": "<p>Really thanks!!!</p>",
      "rawMarkdown": "Really thanks!!!",
      "votes": null
    },
    {
      "id": "520728",
      "postDate": "04/21/2019 16:41:38",
      "content": "<p>And this topic, <a href=\"https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107\">https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107</a></p>",
      "rawMarkdown": "And this topic, https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107",
      "votes": null
    },
    {
      "id": "523455",
      "postDate": "04/26/2019 10:21:56",
      "content": "<p>As a beginner, after read the comments, and still exist doubt about what's Acc@1 and Acc@5？\nThe link <a href=\"https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5\">top-1 &amp; top-5 accuracy</a> topic author gave explain clearly. <br>\nRecord as below and share to more beginnger.</p>\n\n<p><strong>Top@1 *<em>accuracy is the conventional accuracy: the model output (</em>*the one with highest probability</strong>) must be exactly the expected answer.</p>\n\n<p><strong>Top@5</strong> accuracy:  the model output(** top 5th highest probability **) must match the expected answer.</p>\n\n<p>For instance, when you apply a model to recognize  a picture of a cat, and these are the outputs of your neural network:</p>\n\n<p>Tiger: 0.4 <br>\nDog: 0.3 <br>\nCat: 0.1 <br>\nLynx: 0.09 <br>\nLion: 0.08 <br>\nBird: 0.02 <br>\nBear: 0.01  </p>\n\n<p>Using top@1 accuracy, you count this output as wrong, because it predicted a tiger. <br>\nUsing top@5 accuracy, you count this output as correct, because cat is among the top-5 guesses.</p>",
      "rawMarkdown": "As a beginner, after read the comments, and still exist doubt about what's Acc@1 and Acc@5？\nThe link [top-1 &amp; top-5 accuracy](https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5) topic author gave explain clearly.  \nRecord as below and share to more beginnger.\n\n**Top@1 **accuracy is the conventional accuracy: the model output (**the one with highest probability**) must be exactly the expected answer.\n\n**Top@5** accuracy:  the model output(** top 5th highest probability **) must match the expected answer.\n\nFor instance, when you apply a model to recognize  a picture of a cat, and these are the outputs of your neural network:\n\nTiger: 0.4  \nDog: 0.3  \nCat: 0.1  \nLynx: 0.09  \nLion: 0.08  \nBird: 0.02  \nBear: 0.01  \n\nUsing top@1 accuracy, you count this output as wrong, because it predicted a tiger.   \nUsing top@5 accuracy, you count this output as correct, because cat is among the top-5 guesses.",
      "votes": null
    },
    {
      "id": "537430",
      "postDate": "05/27/2019 02:41:48",
      "content": "<p>Thanks!</p>",
      "rawMarkdown": "Thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 518406,
      "author_name": "shnhrtkyk",
      "author_url": "",
      "post_date": "04/17/2019 07:02:30",
      "content": "<p>thanks</p>",
      "votes": null,
      "replies": [
        {
          "id": 518420,
          "author_name": "leonshangguan",
          "author_url": "",
          "post_date": "04/17/2019 07:30:24",
          "content": "<p>Very welcome! I have revised this discussion into table format, which would give you better reading experience!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 518815,
          "author_name": "shnhrtkyk",
          "author_url": "",
          "post_date": "04/17/2019 21:33:47",
          "content": "<p>I was just beginning to study machine learning, \nthis list of networks is very useful information for me.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 518543,
      "author_name": "seefun",
      "author_url": "",
      "post_date": "04/17/2019 12:44:29",
      "content": "<p>Thanks for sharing！If you use pytorch, you can check this <a href=\"https://github.com/Cadene/pretrained-models.pytorch#accuracy-on-validation-set-single-model\">list</a> in <a href=\"https://github.com/Cadene/pretrained-models.pytorch\">pretrainedmodels</a>.</p>",
      "votes": null,
      "replies": [
        {
          "id": 518590,
          "author_name": "jionie",
          "author_url": "",
          "post_date": "04/17/2019 13:49:01",
          "content": "<p>And this kaggle dataset. <a href=\"https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo\">https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 520236,
      "author_name": "dimitreoliveira",
      "author_url": "",
      "post_date": "04/20/2019 13:58:44",
      "content": "<p>Thanks, for sharing, also for anyone using Keras checkout this link: <a href=\"https://keras.io/applications/\">https://keras.io/applications/</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 537430,
          "author_name": "vivaroma",
          "author_url": "",
          "post_date": "05/27/2019 02:41:48",
          "content": "<p>Thanks!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 520243,
      "author_name": "rbarman",
      "author_url": "",
      "post_date": "04/20/2019 14:19:04",
      "content": "<p>Could you explain the Acc@1 vs Acc@5 columns? </p>",
      "votes": null,
      "replies": [
        {
          "id": 520608,
          "author_name": "leonshangguan",
          "author_url": "",
          "post_date": "04/21/2019 12:06:00",
          "content": "<p><a href=\"https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5\">top-1 &amp; top-5 accuracy</a>These means top-1 accuracy and top-5 accuracy separately, I think this website could be helpful</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 520714,
      "author_name": "interneuron",
      "author_url": "",
      "post_date": "04/21/2019 16:14:42",
      "content": "<p>You can find more pretrained and not pretrained models at this excellent repo <a href=\"https://github.com/osmr/imgclsmob\">https://github.com/osmr/imgclsmob</a> </p>\n\n<p>I use it for pytorch, but it has models for Keras and chainer as well, though I have not tried them. </p>",
      "votes": null,
      "replies": [
        {
          "id": 520722,
          "author_name": "jionie",
          "author_url": "",
          "post_date": "04/21/2019 16:35:26",
          "content": "<p>Really thanks!!!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 520728,
      "author_name": "jionie",
      "author_url": "",
      "post_date": "04/21/2019 16:41:38",
      "content": "<p>And this topic, <a href=\"https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107\">https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 523455,
      "author_name": "icekstry",
      "author_url": "",
      "post_date": "04/26/2019 10:21:56",
      "content": "<p>As a beginner, after read the comments, and still exist doubt about what's Acc@1 and Acc@5？\nThe link <a href=\"https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5\">top-1 &amp; top-5 accuracy</a> topic author gave explain clearly. <br>\nRecord as below and share to more beginnger.</p>\n\n<p><strong>Top@1 *<em>accuracy is the conventional accuracy: the model output (</em>*the one with highest probability</strong>) must be exactly the expected answer.</p>\n\n<p><strong>Top@5</strong> accuracy:  the model output(** top 5th highest probability **) must match the expected answer.</p>\n\n<p>For instance, when you apply a model to recognize  a picture of a cat, and these are the outputs of your neural network:</p>\n\n<p>Tiger: 0.4 <br>\nDog: 0.3 <br>\nCat: 0.1 <br>\nLynx: 0.09 <br>\nLion: 0.08 <br>\nBird: 0.02 <br>\nBear: 0.01  </p>\n\n<p>Using top@1 accuracy, you count this output as wrong, because it predicted a tiger. <br>\nUsing top@5 accuracy, you count this output as correct, because cat is among the top-5 guesses.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "518380": "When selecting base model,  I think you can check this table.\n| Model        | Acc@1           | Acc@5  |Time* |\n| ------------- |:-------------:| -----:| -----:|\n| vgg16  | 70.79 | 89.74 |24.95|\n|vgg19 |\t70.89| 89.69|\t24.95|\n|resnet18|68.24|88.49|\t16.07 |\n|resnet34|72.17|90.74|\t17.37|\n|resnet50|74.81|92.38|\t22.62|\n|resnet101|76.58|93.10|33.03|\n|resnet152|76.66|93.08|42.37|\n|resnet50v2|69.73|89.31|19.56|\n|resnet101v2|71.93|90.41|28.80|\n|resnet152v2|72.29|90.61|41.09|\n|resnext50|77.36|93.48|37.57|\n|resnext101|78.48|94.00|60.07|\n|densenet121|74.67|92.04|27.66|\n|densenet169|75.85|92.93|33.71|\n|densenet201\t|77.13|93.43|42.40|\n|inceptionv3|77.55|93.48|38.94|\n|xception|78.87|94.20|42.18|\n|inceptionresnetv2|80.03|94.89|54.77|\n|seresnet18|69.41|88.84|20.19|\n|seresnet34|\t72.60|90.91|22.20|\n|seresnet50|\t76.44|93.02|23.64|\n|seresnet101|\t77.92|94.00|32.55|\n|seresnet152|\t78.34|94.08|47.88\t|\n|seresnext50|\t78.74|94.30|38.29|\n|seresnext101|79.88|94.87|62.80|\n|senet154|81.06|95.24|137.36|\n|nasnetlarge|\t82.12|95.72|116.53|\n|nasnetmobile|74.04|91.54|27.73|\n|mobilenet|70.36|89.39|15.50|\n|mobilenetv2|71.63|90.35|18.31|",
    "518406": "thanks",
    "518420": "Very welcome! I have revised this discussion into table format, which would give you better reading experience!",
    "518543": "Thanks for sharing！If you use pytorch, you can check this [list](https://github.com/Cadene/pretrained-models.pytorch#accuracy-on-validation-set-single-model) in [pretrainedmodels](https://github.com/Cadene/pretrained-models.pytorch).",
    "518590": "And this kaggle dataset. https://www.kaggle.com/igorkrashenyi/pytorch-model-zoo",
    "518815": "I was just beginning to study machine learning, \nthis list of networks is very useful information for me.",
    "520236": "Thanks, for sharing, also for anyone using Keras checkout this link: https://keras.io/applications/",
    "520243": "Could you explain the Acc@1 vs Acc@5 columns?",
    "520608": "[top-1 &amp; top-5 accuracy](https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5)These means top-1 accuracy and top-5 accuracy separately, I think this website could be helpful",
    "520714": "You can find more pretrained and not pretrained models at this excellent repo https://github.com/osmr/imgclsmob \n\nI use it for pytorch, but it has models for Keras and chainer as well, though I have not tried them.",
    "520722": "Really thanks!!!",
    "520728": "And this topic, https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87622#latest-520107",
    "523455": "As a beginner, after read the comments, and still exist doubt about what's Acc@1 and Acc@5？\nThe link [top-1 &amp; top-5 accuracy](https://stackoverflow.com/questions/37668902/evaluation-calculate-top-n-accuracy-top-1-and-top-5) topic author gave explain clearly.  \nRecord as below and share to more beginnger.\n\n**Top@1 **accuracy is the conventional accuracy: the model output (**the one with highest probability**) must be exactly the expected answer.\n\n**Top@5** accuracy:  the model output(** top 5th highest probability **) must match the expected answer.\n\nFor instance, when you apply a model to recognize  a picture of a cat, and these are the outputs of your neural network:\n\nTiger: 0.4  \nDog: 0.3  \nCat: 0.1  \nLynx: 0.09  \nLion: 0.08  \nBird: 0.02  \nBear: 0.01  \n\nUsing top@1 accuracy, you count this output as wrong, because it predicted a tiger.   \nUsing top@5 accuracy, you count this output as correct, because cat is among the top-5 guesses.",
    "537430": "Thanks!"
  },
  "source": "meta"
}