A Convolutional Neural Network Face Recognition Method Based on BiLSTM and Attention Mechanism

Table 1

Evaluation indicators under different batch sizes.

Batch size	Acc	F1-score	Recall	Precision
16	0.9375	0.5714	0.6667	0.5
32	0.9729	0.9230	0.9230	0.9230
64	0.9531	0.8571	0.8571	0.8571
128	0.9609	0.7381	0.8158	0.6739

The bold values indicate that batch size of 32 has the best experimental results, so the batch size used in this paper is 32