Rajiv Kumar, Rishabh Dabral, G. Sivakumar
Unsupervised image-to-image translation is used to transform images from a source domain to generate images in a target domain without using source-target image pairs. Promising results have been obtained for this problem in an adversarial setting using two independent GANs and attention mechanisms. We propose a new method that uses a single shared discriminator between the two GANs, which improves the overall efficacy. We assess the qualitative and quantitative results on image transfiguration, a cross-domain translation task, in a setting where the target domain shares similar semantics to the source domain. Our results indicate that even without adding attention mechanisms, our method performs at par with attention-based methods and generates images of comparable quality.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Image-to-Image Translation | Apples and Oranges | Kernel Inception Distance | 4.4 | Shared discriminator GAN |
| Image-to-Image Translation | Zebra and Horses | Kernel Inception Distance | 5.8 | Shared discriminator GAN |
| Image Generation | Apples and Oranges | Kernel Inception Distance | 4.4 | Shared discriminator GAN |
| Image Generation | Zebra and Horses | Kernel Inception Distance | 5.8 | Shared discriminator GAN |
| 1 Image, 2*2 Stitching | Apples and Oranges | Kernel Inception Distance | 4.4 | Shared discriminator GAN |
| 1 Image, 2*2 Stitching | Zebra and Horses | Kernel Inception Distance | 5.8 | Shared discriminator GAN |