Skip to main content
. 2022 Nov 10;12:19200. doi: 10.1038/s41598-022-21848-3

Figure 6.

Figure 6

The floating-point operations per second (FLOPs) and parameters versus the balanced accuracy of the seven CNN models on FBCG dataset from the fivefold stratified CV (FLOPs in the x-axis depicts the number of operations in billions, while the radius of the circle represents the number of parameters in millions). The EfficientNetV2-B0-21k model scored the highest score (0.9666) with relatively low FLOPs (0.72B) and parameters (7.1 M). The ResNetV1-50 model achieved a low accuracy (0.9253) score with the highest FLOPs (4.1B) and parameters (25.6 M). Most of the CNN models scored average accuracy scores between 0.93 and 0.4. Generally, the average accuracy score is increasing with the FLOPs except for the EfficientNetV2B0 and ResNetV-501 and ResNetV2-50. There is no evidence that larger parameter CNN models (ResNetV1-50 and ResNetV2-50) are more accurate.