A Case Study Evaluating Improved Performance in Image Classification Through Combination of CBAM and ShuffleNetV2 Model

Abstract

The Attention mechanism is a method that focuses attention on important parts or regions in an image while disregarding unimportant areas. Some methods of Attention mechanism include Channel Attention, Spatial Attention, or a combination of Channel and Spatial Attention. CBAM (Convolutional Block Attention Module) is a method that combines both Channel and Spatial Attention. This paper describes a case study on combining CBAM with the ShuffleNetV2 model to evaluate the effectiveness of improving image classification performance. The ShuffleNetV2 model is trained on the CIFAR-10 dataset combined with CBAM for performance evaluation. The training of the ShuffleNetV2 model has been conducted for approximately 40 epochs. The performance evaluation indicates that the ShuffleNetV2 model combined with CBAM yields Precision 89.5%, Recall 89.4%, F1-score 89.4%, Top-5 error 0.003, and Top-1 error 0.106. In comparison, the ShuffleNetV2 model without CBAM achieves Precision, Recall, F1-score, Top-5 error, and Top-1 error of 88.9%, 88.8%, 0.005, 0.112, respectively.