Rajarshi1001 / a-study-in-GANs

GAN model results

1. BasicGAN

Hand-digit generation using basicGAN on MNIST dataset

2. DCGAN

Results on DCGAN on MNIST and CelebA dataset

3. Conditional GAN

Result of cGAN on MNIST, FMNIST and CIFAR

4. infoGAN

Result of infoGAN on MNIST, FMNIST and CIFAR

5. AC-GAN

Result of infoGAN on MNIST, FMNIST and CIFAR

6. Amine GAN

7. Cycle GAN

8. Star GAN

9. SRGAN

Left to Right: (1) Low resolution image (LR) which is also the inputs to our model (2)
High Resolution image (HR) which is ground truth image for evaluating image [4x
resolution of LR] (3) Super resolution image (SR) which is the generated image from the

10. SAGAN

Self Attention module and other stabilizing method applied on DCGAN as proposed in
the paper SAGAN. Above results suggest that our model has overcome the problem of
long range dependencies which can clearly be seen in initial epoch’s result.