Skip to main content

Table 5 Variation of the Average IoU (%) with the layer and threshold used for ResNet-101

From: Deep learning based high-throughput phenotyping of chalkiness in rice exposed to high night temperature

Layer T = 20% T = 30% T = 40% T = 50% T = 60% T = 70% T = 80%
layer1_2_conv2 0.20 9.90 18.41 26.08 37.53 18.55 18.55
layer2_0_conv2 3.81 19.86 31.53 44.90 68.11 18.55 18.55
layer3_1_conv2 1.77 9.59 18.92 28.22 41.59 18.55 18.55
layer4_1_conv3 0.15 10.26 15.43 21.10 29.68 18.55 18.55
  1. The layer is used to generate the heatmaps and the threshold T is used to binarize the heatmaps (e.g., \(T=20\%\) means that only pixels with values at least \(20\%\) of the max pixel value in the image are included in the binary mask). The layers were sampled to include a low-level layer (layer1_2_conv2), a high-level layer (layer4_1_conv3) and two intermediate layers (layer2_0_conv2 and layer3_1_conv2) that showed good results based on a qualitative inspection of the maps. The threshold T is varied from \(20\%\) to \(80\%\) in increments of 10. The best result and the corresponding layer and threshold are highlighted in bold font