{
  "fetched_at": "2026-10-19T23:59:48Z",
  "source_url": "https://www.kaggle.com/competitions/analytics-fundamentals-ii-2022-2-convnets/data",
  "pages": [
    {
      "name": "data-description",
      "content": "## Files\n\n*   **train.csv** - the training set\n*   **test.csv** - the test set\n*   **sample_submission.csv** - a sample submission file in the correct format\n*  **images.zip** - images folder\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "You have been hired by a great startup of the agricultural sector to develop a predictive model that has to determine the localization and type of different vegetables that appear in an image. You are given a dataset containing images and their corresponding annotations. It is guarateed that each image contains only one object of three possible classes, your model should detect which type of object is and where it is located.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "# Evaluation\n\n-  **Custom CNN:**  Create a custom Convolutional Neural Network model as a backbone (feature extractor) and train it from scratch (1.5 points)\n\n-  **Bounding box prediction:** Implement a regression head that is fed with the features extracted from the convolutional backbone. (1 point)\n\n- **Object classification:** Implement a classification head that is fed with the features extracted from the convolutional backbone (1 point)\n\n- **Transfer learning:** Use two pretrained models as backbones and connect the previously created classification and regression heads in order to receive inputs from them. (1 point)\n\n- **Data augmentation:** Compare the use of different data augmentation techniques in order to see their impact in the performance of both tasks (classification and regression) (0.5 points)\n\nBe aware that you must follow a correct evaluation protocol in order to avoid data leakage from the test sets on the data processing steps. Mistakes will be punished (-0.5 points each).\n\nThe evaluation metric for this competition is the [dice coefficient] (https://www.kaggle.com/code/yerramvarun/understanding-dice-coefficient/notebook) for the bounding box regression task and [accuracy](https://developers.google.com/machine-learning/crash-course/classification/accuracy) for the object classification task.\n\n## Submission Format\n\n**For every set in the dataset**, submission files should contain the following columns: \n\n* `filename`, the filename of the image\n* `class`, the predicted class",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": " - Don't cheat!\n - Apply yourself!\n - Have fun!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}