pcvlab/yolov8_ocular_ultrasound_globe_detection

YOLO Ocular Globe Detection

YOLOv8m model trained to detect the ocular globe in ultrasound videos. Part of the ERDES benchmark for classifying ocular ultrasound cases.

Resource	Link
Paper
Dataset
Checkpoints
Code

Classes

ID	Class
0	whole eye

Only class 0 (whole eye) is used for ocular globe detection and cropping. A second class (eyecenter) was annotated during training but is not used in the classification pipeline.

Usage

pip install ultralytics

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model = hf_hub_download("pcvlab/yolo_ocular_globe_detection", "yolo_ocular_globe_detection.pt")
yolo = YOLO(model)
results = yolo.predict("your_video.mp4", conf=0.8)

See individual model cards under PCVLab for the full ocular ultrasound classification pipeline.

Citation

@misc{ozkut2026erdes,
  title={ERDES: A Benchmark Video Dataset for Retinal Detachment and Macular Status Classification in Ocular Ultrasound},
  author={Yasemin Ozkut and Pouyan Navard and Srikar Adhikari and Elaine Situ-LaCasse and Josie Acu{\~n}a and Adrienne Yarnish and Alper Yilmaz},
  year={2026},
  eprint={2508.04735},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2508.04735}
}