Skip to Main Content

Comparing the results of Tables 2 and 3, we can see that the accuracy of AFG on Mini-ImageNet for generic few-shot learning is almost higher than that of the three fine-grained data sets. Especially in the 5-shot setting, the accuracy is 14.7%, 9.5%, and 6.2% higher on Mini-ImageNet than that on Cub Birds, Stanford Dogs, and Stanford Cars, respectively.

Table 3:
The Mean Accuracy (%) on Four Benchmark Data Sets to Verify the Effect of the 3D-Attention Mechanism.
(a) (b) 
Five-Way on Mini-ImageNet Five-Way on Cub Birds 
Model AFG-3D AFG Model AFG-3D AFG 
1-shot 52.70 51.03 ( 1.67) 1-shot 39.74 50.02 ( 10.28) 
5-shot 68.30 69.14 ( 0.84) 5-shot 41.13 54.41 ( 13.28) 
(c) (d) 
Five-Way on Stanford Dogs Five-Way on Stanford Cars 
Model AFG-3D AFG Model AFG-3D AFG 
1-shot 25.80 39.40 ( 13.6) 1-shot 37.20 42.23 ( 5.03) 
5-shot 51.29 59.61 ( 8.32) 5-shot 51.30 62.90 ( 11.6) 
(a) (b) 
Five-Way on Mini-ImageNet Five-Way on Cub Birds 
Model AFG-3D AFG Model AFG-3D AFG 
1-shot 52.70 51.03 ( 1.67) 1-shot 39.74 50.02 ( 10.28) 
5-shot 68.30 69.14 ( 0.84) 5-shot 41.13 54.41 ( 13.28) 
(c) (d) 
Five-Way on Stanford Dogs Five-Way on Stanford Cars 
Model AFG-3D AFG Model AFG-3D AFG 
1-shot 25.80 39.40 ( 13.6) 1-shot 37.20 42.23 ( 5.03) 
5-shot 51.29 59.61 ( 8.32) 5-shot 51.30 62.90 ( 11.6) 

Note: The difference between the two models is marked in bold.

Close Modal

or Create an Account

Close Modal
Close Modal