{
  "fetched_at": "2026-10-19T07:15:39Z",
  "source_url": "https://www.kaggle.com/competitions/find-person-in-the-dark/data",
  "pages": [
    {
      "name": "data-description",
      "content": "## File descriptions\n- ``train.zip``: including all training images and annotations.\n- ``test.zip``: including all testing images.\n- ``train_data.txt``: the training records, the format of bounding boxes is ``classlabel xmin ymin width height``\n- ``test_data.txt``: the testing records.\n- ``evaluate.zip``: including scripts for evaluation.\n- ``submission_example.csv``: an example of submission file.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "## Introduction\nPedestrian detection technology has been developed very mature, but it is still a very big challenge to detect pedestrians in low light conditions. Although infrared light images can assist pedestrian detection, however, in most scenes, only visible light images are available. Therefore, we organized this competition to promote the research of pedestrian detection in low light conditions.\n\n## Dataset\nThe dataset used in this competition contains 15030 visible light images(11782 images for training and 3248 for testing), it is a subset of our previous work: [LLVIP: A Visible-infrared Paired Dataset for Low-light Vision](https://arxiv.org/abs/2108.10831). In low light conditions, even the human eye can hardly distinguish pedestrians. Therefore, our pedestrian marking work is completed with the assistance of infrared light images.\n## License\nOur dataset is made freely available to academic and non-academic entities for non-commercial purposes such as academic research, teaching, scientific publications, or personal experimentation. Permission is granted to use the data given that you agree to our [license terms](https://github.com/bupt-ai-cz/LLVIP/blob/main/Term%20of%20Use%20and%20License.md).\nIf you use this data for your research, please cite our paper ：\n```\n@inproceedings{jia2021llvip,\n  title={LLVIP: A Visible-infrared Paired Dataset for Low-light Vision},\n  author={Jia, Xinyu and Zhu, Chuang and Li, Minzhen and Tang, Wenqi and Zhou, Wenli},\n  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},\n  pages={3496--3504},\n  year={2021}\n}\n```\n## Contact\nIf you encounter any problems please post them in the [Discussion](https://www.kaggle.com/c/find-person-in-the-dark/discussion) column without hesitation, or contact us directly by the following email:\n- shengjie.Liu@bupt.edu.cn\n- czhu@bupt.edu.cn\n## Related\nIf you like our 'Find Person in the Dark' Dataset, you will probably be interested in our other researches, please visit [here](https://github.com/bupt-ai-cz) for more details.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "## Evaluation Metric\nThe evaluation metric for this competition is **Average Precision** at IoU thresholds of 0.5(AP@0.5) and 0.75(AP@0.75). We provide an authoritative [tool](https://github.com/rafaelpadilla/Object-Detection-Metrics) for AP calculation. You can download the file ``evaluate.zip`` in the 'data' column to get the tool.\n### How to use the evaluation tool?\n- Step1: Download and unzip the ``evaluate.zip`` file. \n- Step2: For each test image, output a list of bounding boxes ``classlabel confidencescore xmin ymin width height`` for objects, and save them as a ``.txt`` file. For example:\n\n``190001.txt``:\n```\nperson .98 1000 300 120 260\nperson .89 1200 300 80 220\n```\n``190002.txt``:\n```\nperson .95 260 570 120 300\n```\n- Step3: Put your predicted ``.txt`` files in the folder ``evaluate/detections``.\n- Step4: Run the python script ``pascalvoc.py`` and you will get the AP value of your predicted results.\n```\npython pascalvoc.py --threshold 0.5\n```\n```\npython pascalvoc.py --threshold 0.75\n```\n\n## Submission Format\nYour score will be obtained by calculating the Mean Absolute Error between your AP values and 1.\nThe submission file should contain a header and have the following format:\n```\nID,value\nAP@0.5,0.871\nAP@0.75,0.455\n\n```",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "Only the dataset provided on Kaggle is allowed to be used in the competition, any use of external data is forbidden(e.g. the infrared light images in the LLVIP dataset).",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}