{
  "fetched_at": "2026-10-18T13:54:01Z",
  "source_url": "https://www.kaggle.com/competitions/HyperLeaf2024/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The HyperLeaf2024 dataset contains 2410 hyperspectral images of wheat flag leaves. The goal is to develop methods for whole-image hyperspectral regression. All leaves and associated measurements are collected from the same field with varying plot parameters. There are 1590 training images taken from 24 plots and 820 test images from an additional 12 plots. The goal is to build models using the image data to predict the 8 targets.\n\nSome tips:\n*   The first 3 targets (GrainWeight, Gsw, PhiPS2) have a wide range of possible values and should be treated as a pure regression task.\n*   There are only 3 possible values for the fertilizer content, 0.0, 0.5, 1.0. It may or may not be beneficial to treat this a classification task.\n*   The last 4 targets (cultivars: Heerup, Kvium, Rembrandt, Sheriff) are one-hot encoded and thus for each sample, one of these has a value of 1.0 and the other three are 0.0. Therefore it may also be beneficial to treat this a classification task and include this restriction in your models.\n*   Keep in mind the leaderboard metric is based on mean squared error, so rounding the targets to the nearest value (for fertilizer and cultivars) may reduce your score if the closest value is incorrect. \n\n## Files\n\n*   **images** - folder containing all images in both train an test sets\n*   **train.csv** - the training set, including the ImageId and the targets\n*   **test.csv** - the test set, includes ImageId, targets hidden\n*   **sample_submission.csv** - a sample submission file that simply predicts the mean values for each target\n\n## Columns\n\n*   `ImageId` - ID of the image, corresponds to filename in the images folder\n*   `GrainWeight` - measurement of yield\n*   `Gsw` - stomatal conductance\n*   `PhiPS2` - chlorophyll fluorescence\n*   `Fertilizer` - fertilizer level (0.0, 0.5, or 1.0)\n*   `Heerup` - Heerup cultivar (0.0 or 1.0)\n*   `Kvium` - Kvium cultivar (0.0 or 1.0)\n*   `Rembrandt` - Rembrandt cultivar (0.0 or 1.0)\n*   `Sheriff` - Sheriff cultivar (0.0 or 1.0)\n\n## Wavelengths\n\nThese are the center wavelengths for each channel in the images:\n\n    397.32, 400.20, 403.09, 405.97, 408.85, 411.74, 414.63, 417.52, 420.40, 423.29, 426.19, 429.08,\n    431.97, 434.87, 437.76, 440.66, 443.56, 446.45, 449.35, 452.25, 455.16, 458.06, 460.96, 463.87,\n    466.77, 469.68, 472.59, 475.50, 478.41, 481.32, 484.23, 487.14, 490.06, 492.97, 495.89, 498.80,\n    501.72, 504.64, 507.56, 510.48, 513.40, 516.33, 519.25, 522.18, 525.10, 528.03, 530.96, 533.89,\n    536.82, 539.75, 542.68, 545.62, 548.55, 551.49, 554.43, 557.36, 560.30, 563.24, 566.18, 569.12,\n    572.07, 575.01, 577.96, 580.90, 583.85, 586.80, 589.75, 592.70, 595.65, 598.60, 601.55, 604.51,\n    607.46, 610.42, 613.38, 616.34, 619.30, 622.26, 625.22, 628.18, 631.15, 634.11, 637.08, 640.04,\n    643.01, 645.98, 648.95, 651.92, 654.89, 657.87, 660.84, 663.81, 666.79, 669.77, 672.75, 675.73,\n    678.71, 681.69, 684.67, 687.65, 690.64, 693.62, 696.61, 699.60, 702.58, 705.57, 708.57, 711.56,\n    714.55, 717.54, 720.54, 723.53, 726.53, 729.53, 732.53, 735.53, 738.53, 741.53, 744.53, 747.54,\n    750.54, 753.55, 756.56, 759.56, 762.57, 765.58, 768.60, 771.61, 774.62, 777.64, 780.65, 783.67,\n    786.68, 789.70, 792.72, 795.74, 798.77, 801.79, 804.81, 807.84, 810.86, 813.89, 816.92, 819.95,\n    822.98, 826.01, 829.04, 832.07, 835.11, 838.14, 841.18, 844.22, 847.25, 850.29, 853.33, 856.37,\n    859.42, 862.46, 865.50, 868.55, 871.60, 874.64, 877.69, 880.74, 883.79, 886.84, 889.90, 892.95,\n    896.01, 899.06, 902.12, 905.18, 908.24, 911.30, 914.36, 917.42, 920.48, 923.55, 926.61, 929.68,\n    932.74, 935.81, 938.88, 941.95, 945.02, 948.10, 951.17, 954.24, 957.32, 960.40, 963.47, 966.55,\n    969.63, 972.71, 975.79, 978.88, 981.96, 985.05, 988.13, 991.22, 994.31, 997.40, 1000.49, 1003.58",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Our dataset consists of 2410 hyperspectral images of wheat flag leaves recorded with the portable SpecimIQ (Oulu, Finland) hyperspectral camera. Each image contains a single leaf and has dimensions of 48 x 352 and covers 204 spectral channels in the 400nm-1000nm range. The result is a data cube of shape 48 x 352 x 204. \n\nFor each image there are 4 regression targets: yield (GrainWeight), stomatal conductance (Gsw), chlorophyll fluorescence (PhiPS2), and fertilizer. Additionally, we include a cultivar classification task that has been one-hot encoded (Heerup, Kvium, Rembrandt, Sheriff) and converted into a regression task to facilitate Kaggle's single metric leaderboard. The challenge goal is to develop methods that can be used to predict theses 5 targets as accurately as possible. \n\nGlobally, wheat is one of the main sources of food. Its continued and effective yield in an increasingly harsh climate is essential to providing the nutrients we need in the years to come. As a result, developing methods to research and analyze crops is an important step towards producing effective treatments and mitigating the negative effects of climate change on the global food supply. Hyperspectral imaging is a commonly used technique in the agriculture research to monitor the growth and health of plants. Hyperspectral imaging records both the spatial information, as in a typical RGB image, as well as the spectral information. It works by recording a spectral reflectance curve for each pixel, in turn producing an image with a channel for each wavelength that the camera records. The result is a hyperspectral data cube with two spatial dimensions and one spectral.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated by a standardized weighted mean squared error.\n\nWe chose this metric for the following reasons:\n- Standardization ensures the MSE scores all have the same scale. This ensures comparability between the MSE scores for the individual targets\n- Weighting allows us to weight all four cultivar regression targets such that our average considers them to be a single target.\n- Simplifies intuitive understanding of the metric. Assuming test data is from the same distribution as the training data, blindly guessing the training set mean results in a score of approximately 1.0, with 0.0 being a perfect score.\n\nGiven following parameters:\n\n$$N = \\mathtt{Number\\, of\\, samples}$$\n$$T = \\mathtt{Number\\, of\\, targets}$$\n$$w_j = \\mathtt{Target\\, weight}$$\n$$\\sigma_j= \\mathtt{Target\\, standard\\, deviation\\, (from\\, training\\, set)}$$\n\nWe score all submissions using the following formula:\n\n$$\\frac{1}{N}\\sum_j^T \\frac{w_j}{{\\sigma_j}^2} \\sum_i^N (y_{i,j} - \\hat{y_{i,j}})^2$$\n\nWe using the following target weights:\n\n| target        | GrainWeight | Gsw    | PhiPS2 | Fertilizer | Heerup | Kvium  | Rembrandt | Sheriff |\n|---------------|-------------|--------|--------|------------|--------|--------|-----------|---------|\n| $$w_j$$       | 1.0         | 1.0    | 1.0    | 1.0        | 0.25   | 0.25   | 0.25      | 0.25    |\n\n## Submission File\nFor each ImageId in the test set, you must predict a probability for all 8 target variables. The file should contain a header and have the following format:\n\n    ImageId,GrainWeight,Gsw,PhiPS2,Fertilizer,Heerup,Kvium,Rembrandt,Sheriff\n    00002,6355.478,0.05804877,0.5530334,0.5786164,0.25786164,0.22012578,0.27044025,0.25157234\n    00004,6355.478,0.05804877,0.5530334,0.5786164,0.25786164,0.22012578,0.27044025,0.25157234\n    00005,6355.478,0.05804877,0.5530334,0.5786164,0.25786164,0.22012578,0.27044025,0.25157234\n    etc.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "The task is fairly straightforward. Some of the leaves have been imaged more than once at different exposures. It is not allowed to simply find these repetitions and copy their predictions. These are included to require models to be robust to imaging parameters.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "*Update 19/01/26*:\nAs the challenge has ended and a number of people have now requested ground truth labels for hidden test set for the Public and Private leaderboards, I have uploaded them under the data tab in the solutions.csv for those who are interested in making full use of this dataset.\n\nThe HyperLeaf2024 challenge is part of the FGVC11 Workshop at CVPR 2024. It consists of 2410 hyperspectral images of wheat flag leaves, encompassing multiple strains and fertilizer levels and taken at different exposures and under different sunlight conditions. The challenge is aimed at developing models for fine-grained prediction of wheat field properties, plant health metrics, and yield.\n\nAlthough there already exist publicly available drone and satellite hyperspectral imaging datasets. These typically have the goal of per-pixel image classification (one target per pixel). However, the power of hyperspectral imaging in full image classification and regression (one target per image) is less studied. We hope to promote this use case of hyperspectral imaging in wheat agriculture through this dataset and challenge.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1859215%2Ff0a3e47521ccee9d259a6917058b183b%2Fleaves.png?generation=1709892667975111&alt=media)\n\n# Citation\nIf you use this dataset in your research, please cite this [paper](https://openaccess.thecvf.com/content/CVPR2024W/FGVC11/papers/Laprade_HyperLeaf2024_-_A_Hyperspectral_Imaging_Dataset_for_Classification_and_Regression_CVPRW_2024_paper.pdf):\n\n`William Michael Laprade, Pawel Pieta, Svetlana Kutuzova, Jesper Cairo Westergaard, Mads Nielsen, Svend Christensen, Anders Bjorholm Dahl; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 1234-1243`",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}