{
  "fetched_at": "2026-10-20T08:36:46Z",
  "source_url": "https://www.kaggle.com/competitions/csi-4116-project-3/data",
  "pages": [
    {
      "name": "data-description",
      "content": "Fine-grained image classification (bird species)\nPlease download the dataset: https://drive.google.com/file/d/1l-i9-Z0pNVynAFhyibwAkXSuLrMWiKm2/view?usp=sharing\n\nDataset is organized as follows:\n\n* test_subm.csv\n* train_anno.csv\n* images/\n  - test_0001.jpg\n  …\n  - test_0250.jpg\n  - train_0001.jpg\n  …\n  - train_2750.jpg\n\nThe number of train images: 2750\nThe number of test images: 250",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "* Implement deep learning models (ResNet-34) for bird classification using PyTorch framework\n* Explore data augmentations and hyper-parameter tuning for training\n* Explore inference techniques for improving classification accuracy\n* **Change your team name to your student ID** on the Team tab.\n* Due date: **20th June, 23:55 KST**",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "* Submit *test_subm.csv* file with your predicted class labels\n    + The first column, ‘filename’, should contain the names of each file\n    + The second column, ‘cls’, should contain the predicted class for each file\n* You'll receive a project score based on your highest accuracy",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "* **20 submissions** are allowed for the competition\n    + **-5 points** for extra one submission\n    + - E.g., if you attempted 22 submissions, -10 points\n* Must use **ImageNet-1k pretrained** weights of **ResNet-34** which is available from torchvision\n    + Other than the backbone network, you can flexibly add your customized layers and classifier\n    + ResNet-34 backbone parameters: 21284672 (~21M)\n    + Allowed maximum number of model parameters: 30M",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}