Dataset Viewer
Auto-converted to Parquet Duplicate
format_version
int64
format
string
class_names
list
image_size
list
split
dict
image_count
int64
archive
dict
1
ultralytics-yolo
[ "bird" ]
[ 896, 1600 ]
{ "train": 0.9, "validation": 0.1 }
80,769
{ "path": "data/yolo_birds_1600x896.tar.gz", "size": 17198845261, "sha256": "a3c63269914fd968fdf5f22b75a886a9e2e4fb66b24857dc8c991e6236bf50c3" }

Birds Object Detection 1600x896

A single-class bird object-detection dataset prepared for Ultralytics at a 1600 x 896 camera resolution. The dataset uses a deterministic 90% training and 10% validation split.

Contents

  • data/yolo_birds_1600x896.tar.gz: complete Ultralytics dataset archive.
  • manifest.json: archive size, SHA-256 checksum, and dataset metadata.

After extraction, the archive contains data.yaml and matching image and YOLO label trees:

data.yaml
images/train/
images/val/
labels/train/
labels/val/

Download

from huggingface_hub import hf_hub_download

archive = hf_hub_download(
    repo_id="PBatch23888/birds-object-detection-1600x896",
    repo_type="dataset",
    filename="data/yolo_birds_1600x896.tar.gz",
)

Private repositories require an authenticated Hugging Face account with access to the dataset.

Dataset composition

The data builder combines bird annotations from Open Images, COCO 2017, Pascal VOC 2012, Birdsnap, and NABirds. Open Images group-of and depiction annotations are excluded. All retained annotations are mapped to the single class bird.

Licensing

This is a compilation of multiple upstream datasets. Their respective terms and licenses continue to apply. Review each upstream dataset's license before redistributing or using this compilation.

Downloads last month
23