h3ct0r / cnn_feature_extractor_rpv_2017

This repo has the code of the CNN feature extractor experiments using Tensorflow and Alexnet. The features from layers C1, C5 and FC2, and then tested with several classificators.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CNN features from AlexNet

This repo has the code of the Tensotflow implementation of the AlexNet CNN (Tensorflow > 1.2) and scripts to extract the features from C1, C5 and FC2 layers. The code is ready-to-go, including the trained weights of AlexNet and some images to test.

Install

How to run

  • Put the images on a folder, every class separated by folder name: 000, 001, 002, 003, etc.
  • Create a new config file in JSON format (examples in config/ folder).
  • Run by executing python main.py -c config/*a_config_file.json*

Run with the Iris Dataset

  • python main.py -c config/alexnet_iris.json

Results and plots

All the results and plots are defined on the config file, but generally they are located in the plots/ and results/ folders.

  • plots: are PDF generated files with a normalized confusion matrix using a heatmap color map.
  • results: are JSON files generated with all the relevant information about the experiment: confusion matrix, overall precision, average precision, etc.

Authors

  • Héctor Azpúrua
  • Patricia Almeida
  • Willian Hofner

About

This repo has the code of the CNN feature extractor experiments using Tensorflow and Alexnet. The features from layers C1, C5 and FC2, and then tested with several classificators.

License:GNU General Public License v3.0


Languages

Language:Python 100.0%