{
  "fetched_at": "2026-10-18T13:41:31Z",
  "source_url": "https://www.kaggle.com/competitions/artistic-head-detection/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The `train/` and `test/` folders each contain the image files where the filename is the unique id. \n\n`train_with_boxes.csv` contains an entry for every image in the `train/` folder, where each entry has three columns: `id`, `num_human_heads`, and `boxes`.\nThe columns contain the following info:\n\n`id` corresponds to the filename in the `train/` folder\n\n`num_human_heads` is the number of heads matching the criteria (at least 50x50 and at least partially facing the viewer) in the image\n\n`boxes` is a list of bounding boxes, where each bounding box has the format `(x_min, x_max, y_min, y_max)` that specify the pixel coordinates of the box, measured from the upper left hand corner of the image (e.g. the upper left corner of the box has the coordinates `(x_min, y_min)` and the bottom right corner is `(x_max, y_max)`).\n\n`test.csv` contains an entry for every image in the `test/` folder, and contains only the `id` column without boxes or number of heads.\n\n`sample_submission.csv` shows what your submission should look like. (The numbers in the `guess` column were generated uniformly at random from the integers 0-20 so probably won't be that helpful for you!)\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Computer vision models may not be good art critics, but they certainly can be trained to be good art recognizers. In this competition, we ask you to build an object detection model for identifying heads. The twist? None of the heads are real! Rather than photos of real life, each image is a painting, print, or sketch from public domain artwork. \n\n**This dataset was created by [Scale Rapid](https://scale.com/rapid)**, the fastest annotation platform to help you obtain production-quality labels with no data minimums. We support a wide range of use cases, including classification, transcription, semantic segmentation, polygon image annotations, and more. **To get started, visit [this link](https://scale.com/kaggle) to receive $250 of free labeling credits!**\n\nEach head is at least 50px wide and 50px tall, and labelers were instructed to ignore heads where no part of the face is visible (i.e. only the back of the head is visible to the viewer). Your training data will include bounding box information for all the human heads in the dataset, and for the test set you will be asked to predict the number of human heads without the bounding boxes given to you.\n\nHere is a handy link to Kaggle's [competition documentation](https://www.kaggle.com/docs/competitions), which includes, among other things, instructions on [submitting predictions](https://www.kaggle.com/docs/competitions#making-a-submission).\n\n\n## Acknowledgements\n\nWe thank the New York Metropolitan Museum of Art for supplying the original images for this dataset",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "For simplicity, the evaluation metric for this competition is root mean square error, or [RMSE](https://en.wikipedia.org/wiki/Root-mean-square_deviation). RMSE is a common regression metric that penalizes the deviance from ground truth for both overestimates and underestimates. For each of the ids in the test csv (which correspond to the images in the test folder), you must output a prediction for the number of human heads larger than 50px by 50px and not facing away from the viewer that are present in the image.\n\n## Submission Format\n\n**For every image in the test folder**, submission files should contain two columns: `Id` and `guess`. \n\nThe file should contain a header and have the following format (see `sample_submission.csv` for a more in-depth example):\n```\nId,guess\n034abc,1\n```",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "This is a fun competition for doing some object detection and playing with a dataset available for public use. As such, the rules are simple:\n\n - Honor code!\n - Apply yourself!\n - Have fun!\n\nWe hope you enjoy Scale Rapid!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}