Apple 3D Detection & Localization

Project that train YOLO detection models on depth thresholded images to obtain rough 3D location of each foreground apples.

Results

Model	Speed ^{GTX 1060 (ms)}	Precision	Recall	F1	mAP50	mAP50-95
YOLOv8n	19.1	0.925	0.864	0.893	0.916	0.69
YOLOv8s	22.8	0.921	0.864	0.892	0.926	0.704
YOLOv8m	44.8	0.931	0.852	0.890	0.924	0.711
YOLOv8l	66.0	0.919	0.871	0.894	0.925	0.711
YOLOv8x	96.8	0.914	0.856	0.884	0.919	0.711
YOLOv9c	58.0	0.927	0.867	0.896	0.929	0.718
YOLOv9e	103.0	0.937	0.865	0.900	0.928	0.720

Approach

Example

Install Prerequisites

Install PyTorch>=1.8 based on CUDA version.
Follow guide on pyk4a and install Azure SDK.

Dataset

https://universe.roboflow.com/apple-detection-localization

Models

https://drive.google.com/drive/folders/19eW2avN8TmLwfF5wTqMihWmJXaJe1ctT?usp=drive_link