Research map / Image generation / Editing and style transfer
Exploring Explicit Domain Supervision for Latent Space Disentanglement in Unpaired Image-to-Image Translation
Jianxin Lin, Zhibo Chen, Yingce Xia, Sen Liu, Tao Qin, Jiebo Luo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add image reconstruction loss l_im (ablation) | DosGAN · Facescrub identity translation | top-1 accuracy (%): 55.28 -> 94.01 | Improved |
| add domain-specific feature reconstruction loss l_alpha,f (ablation) | DosGAN · Facescrub identity translation | top-1 accuracy (%): 11.16 -> 94.01 | Improved |
| use discriminator d_feat with l_alpha,r instead of alpha with only l_alpha,f (ablation) | DosGAN · Facescrub identity translation | top-1 accuracy (%): 15.99 -> 94.01 | Improved |
| pre-train and fix the domain classifier alpha instead of joint training (ablation) | DosGAN · Facescrub identity translation | top-1 accuracy (%): 61.21 -> 94.01 | Improved |
| replace StarGAN domain codes with pre-trained domain-specific features | StarGAN · Facescrub identity translation | top-1 accuracy (%): 55.04 -> 94.01 | Improved |
| use DosGAN-c conditional model rather than StarGAN | StarGAN · Facescrub identity translation | top-1 accuracy (%): 55.04 -> 60.79 | Improved |
| replace StarGAN with DosGAN-c on unseen identities | StarGAN · Facescrub unseen 3 identities | top-1 accuracy (%): 6.75 -> 58.46 | Improved |
| replace StarGAN with DosGAN | StarGAN · CelebA hair color translation | classification error rate (%): 19.01% -> 17.56% | Improved |
| replace StarGAN with DosGAN | StarGAN · CelebA gender translation | classification error rate (%): 11.60% -> 11.12% | Improved |
| replace StarGAN with DosGAN | StarGAN · CelebA age translation | classification error rate (%): 25.52% -> 23.79 % | Improved |
| replace StarGAN with DosGAN | StarGAN · Season translation | FID: 81.81 -> 51.14 | Improved |
| use DosGAN-c instead of cd-GAN | cd-GAN · edges->handbags | FID: 84.50 -> 78.67 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StarGAN — StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018 · similarity 0.91)
- Domain-Scalable Unpaired Image Translation via Latent Space Anchoring (TPAMI 2023 · similarity 0.91)
- Conditional Image-to-Image Translation (CVPR 2018 · similarity 0.90)
- Image-to-image translation for cross-domain disentanglement (NeurIPS 2018 · similarity 0.90)
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.90)
- Learning to Transfer: Unsupervised Domain Translation via Meta-Learning (AAAI 2020 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
