Table 4 Performance of deep learning models on test data

From: A detection-screening framework for karez (ancient underground irrigation system) using deep learning and geospatial analysis

Model type

Backbone model

P

R

F1-score

mAP50

mAP50-95

MMDetection

ResNet34

0.788

0.915

0.847

0.803

0.777

MMDetection

ResNet50

0.823

0.913

0.865

0.837

0.833

SSD

ResNet34

0.597

0.810

0.687

0.615

0.299

SSD

ResNet50

0.647

0.740

0.691

0.531

0.228

YoloV3

DarkNet53

0.715

0.894

0.795

0.692

0.525

Faster R-CNN

ResNet34

0.483

0.602

0.536

0.428

0.121

Faster R-CNN

ResNet50

0.555

0.934

0.697

0.724

0.568

Back to article page