https://github.com/devkihyun/image_invert_cnn-tensorflow

Tensorflow implement of 'Understanding Deep Image Representations by Inverting Them'
https://github.com/devkihyun/image_invert_cnn-tensorflow

Last synced: 6 days ago
JSON representation

Tensorflow implement of 'Understanding Deep Image Representations by Inverting Them'

• Host: GitHub
• URL: https://github.com/devkihyun/image_invert_cnn-tensorflow
• Owner: DevKiHyun
• Created: 2019-05-18T07:31:35.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-05-18T08:15:03.000Z (over 5 years ago)
• Last Synced: 2024-07-25T11:10:35.441Z (5 months ago)
• Language: Python
• Size: 50.8 KB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Image_Invert_CNN-Tensorflow (2019/05/18)

## Introduction

I implement a tensorflow model for ["Understanding Deep Image Representations by Inverting Them"](https://arxiv.org/pdf/1412.0035.pdf).

## Environment

- Ubuntu 16.04

- Python 3.6.7

## Depenency

- Numpy

- Opencv2

- matplotlib

- Tensorflow(1.4 <= x <= 1.13)

## Files

- image_invert_cnn.py : main code.

## How to use

### Training

```shell

python image_invert_cnn.py

# Default args: training_epoch = 10000, image_resize = (224,224), learning_rate = 0.01

# You can change args: training_epoch = 5000, image_resize = (448, 448), learning_rate = 0.001

python image_invert_cnn.py --training_epoch 5000 --image_resize (448,448) --learning_rate 0.001

```

## Result

##### Sample image (before resize)

![Alt Text](https://github.com/DevKiHyun/Image_Invert_CNN-Tensorflow/blob/master/content_image.jpg)

##### Result of reconstruct image (From feature vector of block4_conv1)

![Alt Text](https://github.com/DevKiHyun/Image_Invert_CNN-Tensorflow/blob/master/reconstructed_image.jpg)

## Reference

["Understanding Deep Image Representations by Inverting Them"](https://arxiv.org/pdf/1412.0035.pdf).