Single-class detection model for American Bison, fine-tuned from the Ultralytics
YOLO26s backbone (pretrained on COCO).
Images sourced from iNaturalist (research-grade observations).
Bounding boxes generated by MegaDetector v5a (confidence ≥ 0.15).
Split 80 / 10 / 10 train / val / test.
Evaluated on the held-out validation set (best checkpoint).
1from ultralytics import YOLO
2model = YOLO("models/bison/yolo26s_finetuned_bison_by_J.Gong_uwyo_2026-05-28.pt")
3results = model.predict("image.jpg", conf=0.25)
