Continuous and Diverse Image-to-Image Translation via Signed Attribute Vectors
Qi Mao1 Hsin-Ying Lee2 Hung-Yu Tseng3 Jia-Bin Huang4 Siwei Ma1 Ming-Hsuan Yang3
1Peking University 2Snap Research 3University of California, Merced 4Virginia Tech
Reference-guided Comparisons
Below are the comparisons between our approach and other image-to-image and face attribtue editing algorithms evaluated on Male <-> Female Translations.
| Input | Reference | StarGAN-v2 | HomoInterpGAN | FUNIT | Our Result |
|---|---|---|---|---|---|
 |
 |
|
|
|
|
 |
 |
|
|
|
|
 |
 |
|
|
|
|
 |
 |
|
|
|
|
 |
 |
|
|
|
|
 |
 |
|
|
|
|
Latent-guided Comparisons
Below are the comparisons between our approach and other image-to-image and face attribtue editing algorithms evaluated on Male <-> Female Translations.
| Input | StarGAN-v2 | DLOW | RelGAN | Fader Networks | Our Result |
|---|---|---|---|---|---|
 |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|