{
  "fetched_at": "2026-10-19T08:31:03Z",
  "source_url": "https://www.kaggle.com/competitions/data-centric-land-cover-classification-challenge-2/data",
  "pages": [
    {
      "name": "data-description",
      "content": "## What should I expect the data format to be?\n\nThe dataset consists of 5,000 images, each with 256x256 pixels, associated with their noisy labels. The images and labels are PNGs. The labels consist of two classes: background (encoded as 0 in the label data) and building (encoded as 1 in the label data).\n\n## What am I doing?\n\nIn this challenge, you will rank samples based on their levels of label noise. In other words, you'll develop an algorithm that assigns a score to each of the 5,000 training samples based on their level of label noise. This score should then be used to rank the data from least to most noisy.\n\n## Files\n\nAlthough images and labels are in different folders (explained below), the image and corresponding label have the same name. For example, image `training_patches/0_9_2_418_209.png` has a label with exactly the same name, but in the `training_noisy_labels` folder.\n\n    training_patches/ - image files\n    training_noisy_labels/ - image noisy labels\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "In supervised semantic segmentation, obtaining fully accurate and reliable labels is often challenging and costly. Labels are frequently contaminated by noise due to factors such as limited information, human labeling errors, or encoding mistakes. In the real-world, label noise is ubiquitous and can significantly degrade the performance of trained machine learning models. Therefore, it is essential to develop methods capable of automatically identifying noisy samples prior to training, allowing machine learning models to be optimized based on the cleanest and most reliable data available.\n\nThis competition invites you to develop AI algorithms that can accurately rank samples based on their levels of label noise. By effectively ranking the data, one can identify and select the most suitable samples for training machine learning models, thus improving both the efficiency and performance of these models.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated using the [Kendall Tau](https://en.wikipedia.org/wiki/Kendall_rank_correlation_coefficient) between the predicted ranking and the undisclosed one.\n\n## Submission File\n\nAfter generating the ranking of all images (from least to most noisy), you must create a file with a header and the following format:\n\n    id,imageid\n    0,1_26_72_418_418.png\n    1,2_19_53_0_627.png\n    2,2_22_42_627_0.png\n    etc.\n\nThe `id` (DO NOT use `ID` or `Id`) is **NOT** relevant and is **NOT** used in the evaluation process. The `imageid` column should have the name of images (for example, `1_26_72_418_418.png`) **in order from least to most noisy**.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "## One account per participant\n\nYou cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.\n\n## Submission Limits\n\nYou may submit a maximum of 5 entries per day.\n\n## Competition Timeline\n\nCompetition Timeline dates are reflected on the competition’s Overview > Timeline page.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "In this competition, you'll develop an AI-based ranking system that ranks the samples based on their **levels of label noise**. Observe that:\n1. **the main idea of this competition is to rank the images automatically, and NOT to re-annotate the dataset,**\n2. **the ranking should be created using the level of *label noise*, and not the level of image noise.**\n\nTo this end, you'll be given a semantic segmentation dataset composed of 5,000 256x256 images and their corresponding (*noisy*) labels. The labels consist of two classes: background (encoded as 0 in the label data) and building (encoded as 1 in the label data).\n\nSuccess is measured by comparing the submitted ranking with an undisclosed one and calculating the [Kendall Tau score](https://en.wikipedia.org/wiki/Kendall_rank_correlation_coefficient).\n\nFinally, this challenge is part of the [MVEO workshop](https://mveo.github.io/challenge.html) in conjunction with [BMVC 2024](https://bmvc2024.org/).",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}