项目作者： YeongHyeon
项目描述 ：
TensorFlow implementation of Conditional Generative Adversarial Nets (CGAN) with MNIST dataset.
高级语言： Python
项目地址: git://github.com/YeongHyeon/CGAN-TF.git
[TensorFlow] Conditional Generative Adversarial Nets (CGAN)
TensorFlow implementation of Conditional Generative Adversarial Nets (CGAN) with MNIST dataset.
Architecture
Training algorithm
The algorithm for training CGAN [1].
CGAN architecture
The architecture of CGAN [1].
Graph in TensorBoard
Graph of CGAN.
Results
Training Procedure
Loss graph in the training procedure.
Each graph shows loss of the discriminator and loss of the generator respectively.
Test Procedure
From random noise without conditions
|z:2|z:64|z:128|
|:—-:|:—-:|:—-:|
|
|
|
|
From random noise with conditions
|z:2|z:64|z:128|
|:—-:|:—-:|:—-:|
|
|
|
|
Latent space walking with conditions
|Class-0 (z:2)|Class-1 (z:2)|Class-2 (z:2)|Class-3 (z:2)|Class-4 (z:2)|
|:—-:|:—-:|:—-:|:—-:|:—-:|
|
|
|
|
|
|
|Class-5 (z:2)|Class-6 (z:2)|Class-7 (z:2)|Class-8 (z:2)|Class-9 (z:2)|
|:—-:|:—-:|:—-:|:—-:|:—-:|
|
|
|
|
|
|
Environment
- Python 3.7.4
- Tensorflow 1.14.0
- Numpy 1.17.1
- Matplotlib 3.1.1
- Scikit Learn (sklearn) 0.21.3
Reference
[1] Mehdi Mirza and Simon Osindero. (2014). Conditional Generative Adversarial Nets. arXiv preprint arXiv:1411.1784.
