This is a Tensorflow implementation of Cycle-IR approach for content-aware image retargeting. The munuscript is available at: https://ieeexplore.ieee.org/document/8943352.

The retargeting results of our approach on RetargetMe dataset are available at the folder "Our Cycle-IR result".

@ARTICLE{CycleIR_TMM2019, author={W. {Tan} and B. {Yan} and C. {Lin} and X. {Niu}}, journal={IEEE Transactions on Multimedia (TMM'2019)}, title={Cycle-IR: Deep Cyclic Image Retargeting}, year={2019}, volume={}, number={}, pages={1-1}, doi={10.1109/TMM.2019.2959925}, ISSN={1941-0077}, month={},}

This project includes the source code of TensorFlow implementation for our munuscript of "Cycle-IR: Deep Cyclic Image Retargeting". We demonstrate that image retargeting problem can be solved by using a promising way of unsupervised deep learning.

The package requires only a standard computer with GPU and enough RAM to support the operations defined by a user. 
For optimal performance, we recommend a computer with the following specs:
RAM: 32+ GB
CPU: 8+ cores, 3.6+ GHz/core
GPU：GeForce RTX 1080 Ti GPU

numpy 1.15.4
tensorflow 1.6.0
scipy 1.1.0
scikit-learn 0.20.2
scikit-image 0.14.1
opencv-python 3.3.0.10
matplotlib 3.0.2
pillow 5.3.0

A working version of CUDA, python and tensorflow. This should be easy and simple installation. CUDA(https://developer.nvidia.com/cuda-downloads) tensorflow(https://www.tensorflow.org/install/) python(https://www.python.org/downloads/)

Download traning data and put it into the folder of "training data" Download VGG-16 model and put it into the folder of "VGG_MODEL" please be careful of the consistency of these names with the code. These may be some changes to make them consistency.

4.1 run test_CycleIR.py to test the images in the "test_image" folder.

4.2 run train_CycleIR.py to training. The trained model is saved in the ckpt-wgan folder.