Skip to main content
. 2018 Oct 5;12:64. doi: 10.3389/fnbot.2018.00064

Table 1.

Comparison of performance (in terms of MR−4%, the lower the better the top two permanence is bold) on the Caltech validation set using features from different layer(s) of convolutional network.

Conv1 Conv2 Conv3 Conv4 Conv5 Conv4a Conv5a
36.39 30.39 20.32 20.5 31.47 15.46 19.16