ZFNet
ZFNet copied to clipboard
ZFNet (or ZF-Net) implementation by Tensorflow.
In the name of God
ZFNet
This repository contains implementation of ZFNet (Visualizing and Understanding Convolutional Networks) by Tensorflow and the network tested with the CIFAR-10 dataset.
Download the CIFAR-10 dataset
Before train and evaluate the network, you should download the following dataset:
- CIFAR-10 Dataset: https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
Extract the cifar-10-python.tar.gz
file, then your folder structure should be like the following image:
Training CIFAR-10 dataset
To train the network with cifar-10 dataset, type the following command at the command prompt:
python3 ./train.py
Sample images from cifar-10 dataset:
Results
Epoch 0
Train Accuracy = 0.100
Test Accuracy = 0.100
Epoch 1
Train Accuracy = 0.215
Test Accuracy = 0.216
Epoch 2
Train Accuracy = 0.364
Test Accuracy = 0.357
...
Epoch 50
Train Accuracy = 0.994
Test Accuracy = 0.728
...
Epoch 100
Final Train Accuracy = 1.000
Final Test Accuracy = 0.753
Evaluating CIFAR-10 dataset
To evaluate the network with cifar-10 dataset, type the following command at the command prompt:
python3 ./evaluate.py
Dependencies
- Python 3
- numpy
- scipy
- pillow
- tensorflow
Links
- https://arxiv.org/abs/1311.2901
- https://www.cs.toronto.edu/~kriz/cifar.html
- https://github.com/amir-saniyan/ZFNet