Datasets:
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