TABLE 2.

Comparison of the TD-CNN model with patches of size $128\times 128$, and the off-the-shelf Faster R-CNN object detection method with subimage size of $1000\times 1000$ on the tassels dataset.

Method	Complexity	Model training speed (GPU)	F1 score (%)	Output visualization
TD-CNN	Composed of one classification Inception v3 neural network	$1hour$	95.9	Grayscaling
Faster R-CNN	Composed of three neural networks: Feature ResNet50 network, region proposal network (RPN), and detection network	$\approx 16hours$	97.9	Bounding box