Figure - PMC

Skip to main content

An official website of the United States government

Here's how you know

Here's how you know

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

View full-text article in PMC

. 2019 Mar 2;68(6):876–895. doi: 10.1093/sysbio/syz014

Search in PMC
Search in PubMed
View in NLM Catalog
Add to search

© The Author(s) 2019. Published by Oxford University Press, on behalf of the Society of Systematic Biologists.

This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.

PMC Copyright notice

Figure 4. — a) Impact of image size and average versus max global pooling of c5 features on identification accuracy. Data sets are separated by color, and the best result for each data set is indicated with a cross (X). Global average pooling was always better than global max pooling. In general, tended to be the best image size. It was the optimal size for two out of four data sets (global average pooling) or three out of four data sets (global max pooling). b) Impact of feature depth, normalization and feature fusion on identification accuracy. Input images of size and average global pooling was used in all cases. Data set colors and indication of the settings giving the best identification performance as in (a); note that the performance of feature fusion here is slightly different from that in (a) because of minor differences in the protocol between the experiments. Normalization and feature fusion generally improved identification accuracy, although c4 features tended to outperform features from all other individual layers.

Inline graphic — a) Impact of image size and average versus max global pooling of c5 features on identification accuracy. Data sets are separated by color, and the best result for each data set is indicated with a cross (X). Global average pooling was always better than global max pooling. In general, tended to be the best image size. It was the optimal size for two out of four data sets (global average pooling) or three out of four data sets (global max pooling). b) Impact of feature depth, normalization and feature fusion on identification accuracy. Input images of size and average global pooling was used in all cases. Data set colors and indication of the settings giving the best identification performance as in (a); note that the performance of feature fusion here is slightly different from that in (a) because of minor differences in the protocol between the experiments. Normalization and feature fusion generally improved identification accuracy, although c4 features tended to outperform features from all other individual layers.