fruit-counting

Fruit Counting

Description

Fruit detection, instance segmentation, and yield estimation skill for orchard management, supporting apple, citrus, and other fruit counting using YOLO and SAM combined approaches.

When to Use

• User needs to count fruits in orchard images
• Need to estimate crop yield from image data
• Need to perform instance segmentation on fruit images
• Working with MinneApple or similar fruit datasets

Prerequisites

pip install torch torchvision ultralytics segment-anything timm

Supported Datasets

Dataset	Task	Images	Fruit
MinneApple	Detection, segmentation, counting	1,000+	Apple

Quick Start

from agri_cv_research.models import YOLODetector
from agri_cv_research.datasets import MinneAppleDataset

dataset = MinneAppleDataset(root="./data/minneapple", split="train", download=True)

detector = YOLODetector(model_size="m", task="detect", num_classes=1)
detector.train(dataset, epochs=100, batch_size=16, imgsz=640)

predictions = detector.predict(test_image)
print(f"Apple count: {len(predictions['boxes'])}")

Code Examples

Combined SAM + YOLO for Precise Counting

from agri_cv_research.models import YOLODetector, SAMSegmentor

detector = YOLODetector(model_size="m", task="detect")
detector.train(dataset, epochs=100)

# YOLO provides bounding boxes, SAM refines segmentation
segmentor = SAMSegmentor(model_id="sam_vit_b")
boxes = detector.predict(image)["boxes"]
segmented_fruits = segmentor.predict(image, boxes=boxes)