rf-detrNano-uavs-detectv2

UAVS model trained on all verified object classes, collapsed to only one class: - object.

rf-detr size Nano

Peter Robicheaux, James Gallagher, Joseph Nelson, Isaac Robinson. (Mar 20, 2025). RF-DETR: A SOTA Real-Time Object Detection Model. Roboflow Blog: https://blog.roboflow.com/rf-detr/

Classes collapsed to object class:

Batray Bird Boat Cement_Ship Egregia Fish Jelly Kayak Kelp Mola Mooring_Buoy Otter Person Pinniped Secci_Disc Shark Surfboard Velella_velella Velella_velella_raft Whale

Acts as object detector to feed detection stage of a pipeline

same dataset, same split as uavs-yolo11x-detectv3. https://huggingface.co/MBARI-org/yolo11x-uavs-detectv3

20 epochs

Class all

Total original images in dataset: 2935

(Original images are 7,952 by 5,304 pixels, 42.4 megapixels.)

Total original localizations in dataset: 33482

Images, localizations transformed to 640 x 640 pixels, 50% overlap

Transformed images: 92703

Transformed localizations: 175178

Train/val/test split 85%/10%/5%

Train 78728 images