{
  "id": 132722,
  "title": "Cutmix/Mixup Reports",
  "url": "/competitions/bengaliai-cv19/discussion/132722",
  "author_name": "",
  "post_date": "2020-02-27T13:26:24.476381200Z",
  "votes": 17,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Setup:\n<code>\nImage: 64x64 grayscale images\nValidation Split: 0.15\nModel: Resnet18\nEpochs: 30 epochs (Approximately 30 minutes to train on rtx2070 super)\nBatch Size: 256 \nOptimizer: Adam\nLearning Rate: 0.001\n</code>\nResults (Validation Grapheme Recall):\n<code>\nBaseline (No augmentation): 0.88502\nMixup (Alpha=0.2): 0.91040\nCutmix (ratio between 0.2 ~ 0.8): 0.91810\n</code></p>",
  "messages": [
    {
      "id": "758140",
      "postDate": "02/27/2020 13:26:24",
      "content": "<p>Setup:\n<code>\nImage: 64x64 grayscale images\nValidation Split: 0.15\nModel: Resnet18\nEpochs: 30 epochs (Approximately 30 minutes to train on rtx2070 super)\nBatch Size: 256 \nOptimizer: Adam\nLearning Rate: 0.001\n</code>\nResults (Validation Grapheme Recall):\n<code>\nBaseline (No augmentation): 0.88502\nMixup (Alpha=0.2): 0.91040\nCutmix (ratio between 0.2 ~ 0.8): 0.91810\n</code></p>",
      "rawMarkdown": "Setup:\n```\nImage: 64x64 grayscale images\nValidation Split: 0.15\nModel: Resnet18\nEpochs: 30 epochs (Approximately 30 minutes to train on rtx2070 super)\nBatch Size: 256 \nOptimizer: Adam\nLearning Rate: 0.001\n```\nResults (Validation Grapheme Recall):\n```\nBaseline (No augmentation): 0.88502\nMixup (Alpha=0.2): 0.91040\nCutmix (ratio between 0.2 ~ 0.8): 0.91810\n```",
      "votes": null
    },
    {
      "id": "759020",
      "postDate": "02/28/2020 13:06:28",
      "content": "<p>Thanks a lot for sharing!  👋 \nThis type of small image size with a fast model is the way to go for beginners like me. I can get feedback of model adjustment right away.</p>",
      "rawMarkdown": "Thanks a lot for sharing!  👋 \nThis type of small image size with a fast model is the way to go for beginners like me. I can get feedback of model adjustment right away.",
      "votes": null
    },
    {
      "id": "759494",
      "postDate": "02/29/2020 05:33:01",
      "content": "<p>Yes. The reason why I choose resnet18 is it's very easy to train and sensitive to change.</p>",
      "rawMarkdown": "Yes. The reason why I choose resnet18 is it's very easy to train and sensitive to change.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 759020,
      "author_name": "barnwellguy",
      "author_url": "",
      "post_date": "02/28/2020 13:06:28",
      "content": "<p>Thanks a lot for sharing!  👋 \nThis type of small image size with a fast model is the way to go for beginners like me. I can get feedback of model adjustment right away.</p>",
      "votes": null,
      "replies": [
        {
          "id": 759494,
          "author_name": "quandapro",
          "author_url": "",
          "post_date": "02/29/2020 05:33:01",
          "content": "<p>Yes. The reason why I choose resnet18 is it's very easy to train and sensitive to change.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "758140": "Setup:\n```\nImage: 64x64 grayscale images\nValidation Split: 0.15\nModel: Resnet18\nEpochs: 30 epochs (Approximately 30 minutes to train on rtx2070 super)\nBatch Size: 256 \nOptimizer: Adam\nLearning Rate: 0.001\n```\nResults (Validation Grapheme Recall):\n```\nBaseline (No augmentation): 0.88502\nMixup (Alpha=0.2): 0.91040\nCutmix (ratio between 0.2 ~ 0.8): 0.91810\n```",
    "759020": "Thanks a lot for sharing!  👋 \nThis type of small image size with a fast model is the way to go for beginners like me. I can get feedback of model adjustment right away.",
    "759494": "Yes. The reason why I choose resnet18 is it's very easy to train and sensitive to change."
  },
  "source": "meta"
}