Confusion over Figure 2 and Table 1

[Fay-why]

Thanks for your work! I noticed the information in Figure 2: the model correctly predicts the high-frequency reconstructed image. But the Table 1 shows that the 'HFC' has bad gerneralization than 'LFC', some of test acc of HFC is 0.1281. Is that a contradiction? Please correct me if I'm wrong.