Spatial-Channel Attention-Based Class Activation Mapping for Interpreting CNN-Based Image Classification Models

Table 1

Error rates (%) [23] of four different networks on ImageNet dataset and feature map size of the last convolutional layer.

Network	Top-1 error	Top-5 error	Feature map size
SqueezeNet	41.81	19.38	13 × 13 × 1000
ResNet-18	30.24	10.92	7 × 7 × 512
ResNet-50	23.85	7.13	7 × 7 × 2048
DenseNet-161	22.35	6.20	7 × 7 × 2208