{"success":true,"data":{"author":"Imageomics Institute (Kholiavchenko et al.)","bucketAccessType":"public","bucketName":"imageomics/KABR","bucketPrefix":"","category":"Life Sciences","cloudProvider":"huggingface","createdAt":"2026-11-14T01:01:20.110Z","description":"Drone video of Kenyan wildlife annotated for animal behavior recognition, covering giraffes, plains zebras and Grevy's zebras filmed at the Mpala Research Centre in January 2023. More than 10 hours of footage at 5472 by 3078 pixels and 29.97 frames per second were labelled by a team of 10 overseen by an expert zoologist, with eight classes: walk, trot, run, graze, browse, head up, auto-groom and occluded. Animals were detected with YOLOv8 and tracked with SORT, and 400 by 300 pixel mini-scenes centred on each animal were cut out to reduce the effect of drone motion. The dataset holds 1,139,893 frames: 488,638 of Grevy's zebras, 492,507 of plains zebras and 158,748 of giraffes. Listing the repository (measured 2026-11-13) shows KABR/ with annotation/ (train and val CSV files and class distribution), configs/ and dataset/image/, where image sequences are stored as 2 GB zip parts per species with MD5 files, plus download.py; about 233 GB in the repository. Concatenate the parts and unzip (or run download.py, which needs about 116 GB free), then read frames with OpenCV or PIL; image2video.py rebuilds clips. Benchmarks with I3D, SlowFast and X3D are reported. Released under CC0-1.0; citation of the WACV 2024 workshop paper is requested.","displayTitle":"KABR: Kenyan Animal Behavior Recognition from Drone Videos","doi":"10.57967/hf/1010","id":"a5a996f9-8c1a-54ef-8ba9-3368932c0547","license":"CC0-1.0","licenseEvidenceUrl":"https://huggingface.co/datasets/imageomics/KABR/raw/main/README.md","objectCountEstimate":45,"probeStatus":"ok","referenceUrl":"https://openaccess.thecvf.com/content/WACV2024W/CV4Smalls/papers/Kholiavchenko_KABR_In-Situ_Dataset_for_Kenyan_Animal_Behavior_Recognition_From_Drone_WACVW_2024_paper.pdf","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"imageomics/KABR\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"imageomics/KABR\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=imageomics, license:cc0-1.0, size_categories:1M<n<10M, format:csv, downloads=807"],"format":"mixed","generatedAt":"2026-11-14T01:01:26.439Z","layout":{"example":"KABR/annotation/train.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"original_vido_id video_id frame_id path labels\nZG0001.1 1 1 ZG0001.1/1.jpg 0\nZG0001.1 1 2 ZG0001.1/2.jpg 0\nZG0001.1 1 3 ZG0001.1/3.jpg 0\nZG0001.1 1 4 ZG0001.1/4.jpg 0\nZG0001.1 1 5 ZG0001.1/5.jpg 0\nZG0001.1 1 6 ZG0001.1/6.jpg 0\nZG0001.1 1 7 ZG0001.1/7.jpg 0\nZG0001.1 1 8 ZG0001.1/8.jpg 0\nZG0001.1 1 9 ZG0001.1/9.jpg 0\nZG0001.1 1 10 ZG0001.1/10.jpg 0\nZG0001.1 1 11 ZG0001.1/11.jpg 0\nZG0001.1 1 12 ZG0001.1/12.jpg 0\nZG0001.1 1 13 ZG0001.1/13.jpg 0\nZG0001.1 1 14 ZG0001.1/14.jpg 0\nZG0001.1 1 15 ZG0001.1/15.jpg 0\nZG0","encoding":"utf8","key":"KABR/annotation/train.csv","truncatedAt":512}],"sampleObjects":["KABR/annotation/classes.json","KABR/annotation/distribution.xlsx","KABR/annotation/train.csv","KABR/annotation/val.csv","KABR/configs/I3D.yaml"],"sidecars":[{"bytes":"KABR: High-Quality Dataset for Kenyan Animal Behavior Recognition from Drone Videos\n\n---------------------------------------------------------------------------------------------------\n\nWe present a novel high-quality dataset for animal behavior recognition from drone videos. The dataset is focused on Kenyan wildlife and contains behaviors of giraffes, plains zebras, and Grevy's zebras. The dataset consists of more than 10 hours of annotated videos, and it includes eight different classes, encompassing seven types of animal behavior and an additional category for occluded instances. In the annotation process for this dataset, a team of 10 people was involved, with an expert zoologist overseeing the process. Each behavior was labeled based on its distinctive features, using a standardized set of criteria to ensure consistency and accuracy across the annotations. The dataset was collected using drones that flew over the animals in the Mpala Research Centre in Kenya, providing high-quality video footage of the animal's natural behaviors. We believe that this dataset will be a valuable resource for researchers working on animal behavior recognition, as it provides a diverse and high-quality set of annotated videos that can be used for evaluating deep learning models. Additionally, the dataset can be used to study the behavior patterns of Kenyan animals and can help to inform conservation efforts and wildlife management strategies. We provide a detailed description of the dataset and its annotation process, along with some initial experiments on the dataset using conventional deep learning models. The results demonstrate the effectiveness of the dataset for animal behavior recognition and highlight the potential for further research in this area.\n\n---------------------------------------------------------------------------------------------------\n\nThe KABR dataset follows the Charades format. The Charades format:\n\nKABR\n    /images\n        /video_1\n            /image_1.jpg\n            /image_2.jpg\n            ...\n            /image_n.jpg\n        /video_2\n            /image_1.jpg\n            /image_2.jpg\n            ...\n            /image_n.jpg\n        ...\n        /video_n\n            /image_1.jpg\n            /image_2.jpg\n            /image_3.jpg\n            ...\n            /image_n.jpg\n    /annotation\n        /classes.json\n        /train.csv\n        /val.csv\n\nThe dataset can be directly loaded and processed by the SlowFast (https://github.com/facebookresearch/SlowFast) framework.\n\n---------------------------------------------------------------------------------------------------\n\nNaming:\nG0XXX.X - Giraffes\nZP0XXX.X - Plains Zebras\nZG0XXX.X - Grevy's Zebras\n\n---------------------------------------------------------------------------------------------------\n\nInformation:\n\nKABR/configs: examples of SlowFast framework configs.\nKABR/annotation/distribution.xlsx: distribution of classes for all videos.\n\n---------------------------------------------------------------------------------------------------\n\nScripts:\nimage2video.py: Encode image sequences into the original video. For example, [image/G0067.1, image/G0067.2, ..., image/G0067.24] will be encoded into video/G0067.mp4.\nimage2visual.py: Encode image sequences into the original video with corresponding annotations. For example, [image/G0067.1, image/G0067.2, ..., image/G0067.24] will be encoded into visual/G0067.mp4.","encoding":"utf8","key":"KABR/README.txt","name":"README.txt","size":3420,"truncated":false},{"bytes":"---\nlicense: cc0-1.0\ntask_categories:\n- video-classification\ntags:\n- zebra\n- giraffe\n- plains zebra\n- Grevy's zebra\n- video\n- animal behavior\n- behavior recognition\n- annotation\n- annotated video\n- conservation\n- drone\n- UAV\n- imbalanced\n- Kenya\n- Mpala Research Centre\npretty_name: >-\n  KABR: In-Situ Dataset for Kenyan Animal Behavior Recognition from Drone\n  Videos\ndescription: \"Initial KABR project release, contains drone video clips (mini-scenes) of giraffes, plains zebras, and Grevy's zebras with behavior labels from a subset of videos collected at the Mpala Research Centre in January 2023.\"\nsize_categories:\n- 1M<n<10M\n---\n# Dataset Card for KABR: In-Situ Dataset for Kenyan Animal Behavior Recognition from Drone Videos\n\n## Dataset Description\n\n- **Homepage:** [KABR Mini-Scene Site](https://kabrdata.xyz/)\n- **Project Page:** [KABR Site](https://imageomics.github.io/KABR/)\n- **Repository:** https://github.com/Imageomics/kabr-tools\n- **Paper:** https://openaccess.thecvf.com/content/WACV2024W/CV4Smalls/papers/Kholiavchenko_KABR_In-Situ_Dataset_for_Kenyan_Animal_Behavior_Recognition_From_Drone_WACVW_2024_paper.pdf\n\n### Dataset Summary\n\nWe present a novel high-quality dataset for animal behavior recognition from drone videos. \nThe dataset is focused on Kenyan wildlife and contains behaviors of giraffes, plains zebras, and Grevy's zebras. \nThe dataset consists of more than 10 hours of annotated videos, and it includes eight different classes, encompassing seven types of animal behavior and an additional category for occluded instances. \nIn the annotation process for this dataset, a team of 10 people was involved, with an expert zoologist overseeing the process. \nEach behavior was labeled based on its distinctive features, using a standardized set of criteria to ensure consistency and accuracy across the annotations. \nThe dataset was collected using drones that flew over the animals in the [Mpala Research Centre](https://mpala.org/) in Kenya, providing high-quality video footage of the animal's natural behaviors.\nThe drone footage is captured at a resolution of 5472 x 3078 pixels, and the videos were recorded at a frame rate of 29.97 frames per second.\n\n\n<!--This dataset card aims to be a base template for new datasets. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md?plain=1).-->\n\n### Supported Tasks and Leaderboards\n\nThe results of our evaluation using I3D, SlowFast, and X3D architectures are given in the table below. For each one, the model was trained for 120 epochs with batch size of 5. For more information on these results, see our [paper](coming soon).\n\n| Method | All | Giraffes | Plains Zebras | Grevy’s Zebras |\n| ---- | ---- | ---- | ---- | ---- |\n| I3D (16x5)  |  53.41 | 61.82 | 58.75 | 46.73 |\n| SlowFast (16x5, 4x5) | 52.92 | 61.15 | 60.60 | 47.42 |\n| X3D (16x5) | 61.9 | 65.1 | 63.11 | 51.16 |\n\n### Languages\n\nEnglish\n\n## Dataset Structure\n\nUnder `KABR/dataset/image/`, the data has been archived into `.zip` files, which are split into 2GB files. These must be recombined and extracted. \nAfter cloning and navigating into the repository, you can use the following commands to do the reconstruction:\n```bash\ncd KABR/dataset/image/\ncat giraffes_part_* > giraffes.zip\nmd5sum giraffes.zip # Compare this to what's shown with `cat giraffes_md5.txt`\nunzip giraffes.zip\nrm -rf giraffes_part_*\n\n# Similarly for `zebras_grevys_part_*` and `zebras_plains_part_*`\n```\n\nAlternatively, there is a download script, `download.py`, which allows a download of the entire dataset in its established format without requiring one to clone the repository (cloning requires _at least_ double the size of the dataset to store). To proceed with this approach, download `download.py` to the system where you want to access the data.\nThen, in the same directory as the script, run the following to begin the download:\n```\npip install requests\npython download.py\n```\n\nThis script then downloads all the files present in the repository (without making a","encoding":"utf8","key":"README.md","name":"README.md","size":14136,"truncated":true}],"sourceObjectKey":"KABR/annotation/train.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/imageomics/KABR/KABR/annotation/train.csv\")\ndf.head()","starterSamplePath":"KABR/annotation/distribution.xlsx","totalObjects":45},"summary":"Drone video of Kenyan wildlife annotated for animal behavior recognition, covering giraffes, plains zebras and Grevy's zebras filmed at the Mpala Research Centre in January 2023.","title":"KABR: Kenyan Animal Behavior Recognition from Drone Videos","updatedAt":"2026-11-14T02:37:46.076Z","visibility":"public","websiteUrl":"https://imageomics.github.io/KABR/"}}