Awesome

Bridging the Gap between Label- and Reference based Synthesis(ICCV 2021)

Tensorflow implementation of Bridging the Gap between Label- and Reference-based Synthesis in Multi-attribute Image-to-Image Translation.

Overview architecture

<p align="center"> <img src="./arch.png" width="70%"><br><center></center></p>

Experiment Results

CelebA

<p align="center"> <img src="./l.png" width="95%"><br><center></center></p> <p align="center"> <img src="./r1.png" width="50%"><br><center></center></p> <p align="center"> <img src="./r2.png" width="50%"><br><center></center></p>

Preparation

Prerequisites
- Tensorflow 1.15
- Python 2.x with matplotlib, numpy and scipy
Dataset
- CelebA
- Images should be placed in ./data/CelebA/img_align_celeba
- tfrecords file should be placed in ./data/CelebA/celeba_tfrecords

Quick Start

Exemplar commands are listed here for a quick start.

dataset

prepare dataset to product tfrecords file
```
python data_noise_in.py 
```

Training

To train with size of 128 X 128

python train_arch9.py --experiment_name "file_name" --gpu "gpu_num"

Testing

Example of test

python  test_arch9.py --experiment_name "file_name" --gpu "gpu_num"

Citation

If this work is useful for your research, please consider citing: