Table 8 Double-Head R-CNN performance on the Augmented PCB Defect dataset.

From: Attentive context and semantic enhancement mechanism for printed circuit board defect detection with two-stage and multi-stage object detectors

Method

mAP

\(D_1\)

\(D_2\)

\(D_3\)

\(D_4\)

\(D_5\)

\(D_6\)

Double-Head R-CNN w/Resnet50 (baseline9a)

97.0

95.4

97.3

96.1

96.5

97.3

99.5

Double-Head R-CNN w/Resnet101 (baseline9b)

98.0

97.8

97.5

96.1

98.1

99.2

98.0

Double-Head R-CNN w/Resnet50+ ACASEM

98.6

98.0

98.3

98.8

98.7

98.6

99.4

Double-Head R-CNN w/Resnet101+ ACASEM

98.2

98.5

97.8

97.9

97.9

98.2

99.1

  1. \(D_1\) stands for open-circuit, \(D_2\) : short, \(D_3\) : mousebite, \(D_4\) : spur, \(D_5\) : spurious_copper and \(D_6\) : missing_hole defects.
  2. Significant values are in bold.
Back to article page