Table 3.

Configurations of CA-Conv-LSTM architectures.

Model	CNN model	Class Attention Map	Bi.
CA-VGG-LSTM	VGG-16	$28\times 28\times N$
CA-VGG-BiLSTM	VGG-16	$28\times 28\times N$	✓
CA-GoogLeNet-LSTM	Inception-v3	$17\times 17\times N$
CA-GoogLeNet-BiLSTM	Inception-v3	$17\times 17\times N$	✓
CA-ResNet-LSTM	ResNet-50	$28\times 28\times N$
CA-ResNet-BiLSTM	ResNet-50	$28\times 28\times N$	✓

N indicates the number of classes in the dataset.

Bi. indicates whether the model is bidirectional or not.