LibreYOLO/coco-val2017

COCO val2017 (YOLO format)

coco-val2017.zip holds the 5,000 COCO val2017 images in coco-val2017/images/val2017 and 4,952 YOLO-format label files (80 COCO classes, ids 0 to 79) in coco-val2017/labels/val2017. Images without objects have no label file. There are no training images, so this set is for validation and benchmarking.

Usage with LibreYOLO

import importlib.resources
import zipfile

import yaml
from huggingface_hub import hf_hub_download
from libreyolo import LibreYOLO

zip_path = hf_hub_download(
    repo_id="LibreYOLO/coco-val2017",
    filename="coco-val2017.zip",
    repo_type="dataset",
)
zipfile.ZipFile(zip_path).extractall(".")

# Reuse the COCO class names that ship with LibreYOLO.
coco = importlib.resources.files("libreyolo") / "config/datasets/coco.yaml"
names = yaml.safe_load(coco.read_text())["names"]
cfg = {"train": "images/val2017", "val": "images/val2017", "nc": 80, "names": names}
yaml.safe_dump(cfg, open("coco-val2017/data.yaml", "w"))

model = LibreYOLO("LibreYOLOXn.pt")
model.val(data="coco-val2017/data.yaml")